ue-mcp 1.2.4 → 1.3.0

package/dist/ue-mcp.default.yml

@@ -98,11 +98,9 @@ tasks:
     options:
       method: set_config
   project.build:
-    class_path: ue-mcp.bridge
+    class_path: project.build
     group: project
-    description: "Build C++ project. Params: configuration?, platform?, clean?"
-    options:
-      method: build_project
+    description: "Build the project's C++ out of process with UnrealBuildTool. Works with the editor STOPPED, which a full rebuild requires (UBT cannot link while an editor holds the module DLLs). Blocks until the build finishes and returns the compiler output. Params: configuration? (default Development), platform? (default the host platform), clean? (#958)"
   project.generate_project_files:
     class_path: ue-mcp.bridge
     group: project
@@ -145,6 +143,12 @@ tasks:
     description: Report Live Coding availability/state (available, started, enabledForSession, compiling). Helps choose between live_coding_compile and build_project.
     options:
       method: live_coding_status
+  project.resolve_collision_profile:
+    class_path: ue-mcp.bridge
+    group: project
+    description: "Read one collision profile's resolved per-channel responses: collisionEnabled, objectType, and every channel with Block/Overlap/Ignore. This is the project-side half of blueprint(get_component_collision) (#925): a component's ResponseArray only lists the channels it OVERRIDES, so the profile is where an inherited response actually comes from. Project trace and object channels appear under their configured names, with enumName (ECC_GameTraceChannel1) alongside so a caller can key on something stable. By default the eight engine channels plus every channel the project configured are returned; includeAllChannels=true adds the unused slots. channel narrows it to one. A profile that does not exist lists the ones that do. Params: profileName, channel?, includeAllChannels?"
+    options:
+      method: resolve_collision_profile
   project.write_cpp_file:
     class_path: project.write_cpp_file
     group: project
@@ -220,9 +224,15 @@ tasks:
   asset.bulk_rename:
     class_path: ue-mcp.bridge
     group: asset
-    description: "Batched rename using IAssetTools::RenameAssets - single transaction with one redirector-fixup pass (matches Content Browser drag). Use this over looped rename for scene-referenced assets. World assets are rejected (status=rejected_world); use rename_asset which handles WP externals atomically (#409). Params: renames[] where each entry is {sourcePath, destinationPath} OR {assetPath, newName}."
+    description: "Batched rename using IAssetTools::RenameAssets - single transaction with one redirector-fixup pass (matches Content Browser drag). Use this over looped rename for scene-referenced assets. World assets are rejected (status=rejected_world); use rename_asset which handles WP externals atomically (#409). The fix-up pass cannot see a referencer that is not loaded, so each renamed item reports redirectorLeft and the result carries redirectorsLeft/redirectorsRemoved plus redirectorPackages, which feeds straight into fixup_redirectors (#908). Params: renames[] where each entry is {sourcePath, destinationPath} OR {assetPath, newName}."
     options:
       method: bulk_rename_assets
+  asset.fixup_redirectors:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Clean up ObjectRedirectors left at the old paths after a move, bounded to exactly the packages you name. Preflights each redirector, its destination, and its hard AND soft referencers, then reports packagesToLoad/packagesToSave before touching anything; dryRun=true stops there. The real run loads only those referencers (a soft reference in an unloaded package is what the rename's own fix-up pass misses), rewrites them, saves them, RE-QUERIES the asset registry, and deletes a redirector only when nothing references it any more, reporting kept vs deleted per package with the referencers that survived. Naming a whole content root is refused unless allowProjectWide=true, so this can never turn into a project-wide resave. Protected mounts are refused outright. dryRun is the way to preview: save=false only skips the explicit save pass, because the editor's own fix-up writes what it can regardless. Params: paths[] (redirector packages or the folders holding them), dryRun? (default false), save? (default true), allowProjectWide? (default false) (#908)"
+    options:
+      method: fixup_redirectors
   asset.move:
     class_path: ue-mcp.bridge
     group: asset
@@ -232,13 +242,13 @@ tasks:
   asset.delete:
     class_path: ue-mcp.bridge
     group: asset
-    description: "Delete asset. On failure returns reason (open_in_editor / has_referencers / in_memory_referenced / package_read_only / package_dirty / unknown) plus referencers, inMemoryReferencers, packageReadOnly, packageDirty diagnostics (#601). Pass force=true to auto-close any open asset editors before deleting (#278). Params: assetPath, force?"
+    description: "Delete asset. force defaults to false and is a REAL guard: the bridge asks the Asset Registry for referencers first and refuses with success=false, reason='has_referencers' and the full referencers list rather than destroying an asset other packages point at (#976). force=true takes the force-delete path, which also auto-closes open asset editors (#278). On a delete the editor attempted and could not finish, the reason is open_in_editor / has_referencers / in_memory_referenced / package_read_only / package_dirty / unknown, with referencers, inMemoryReferencers, packageReadOnly and packageDirty diagnostics (#601). Params: assetPath, force?"
     options:
       method: delete_asset
   asset.delete_batch:
     class_path: ue-mcp.bridge
     group: asset
-    description: "Batch-delete assets. Per-path status (deleted/absent/failed) plus reason+referencers on failed entries (#278). Params: assetPaths[], force?"
+    description: "Batch-delete assets. Per-path status is deleted | absent | protected | refused | failed, plus reason+referencers on refused and failed entries. A referenced asset is refused per entry with force=false and the rest of the batch still runs, so one guarded asset never hides behind a bare failed count (#976, #278). Result carries deleted/absent/refused/failed/protected/total. Params: assetPaths[], force?"
     options:
       method: delete_asset_batch
   asset.create_data_asset:
@@ -253,6 +263,12 @@ tasks:
     description: "Create an asset of ANY concrete UObject class (not just UDataAsset) - physical-material subclasses, curves, settings objects. className accepts the C++ spelling with or without the A/U/F/E prefix, a /Script/Module.ClassName path, or a loaded class name (#823). Params: name, className, packagePath?, properties? (key/value map), onConflict? (skip|replace|rename). Actors/components and specialized assets (Blueprint/Material) have dedicated actions (#726)"
     options:
       method: create_asset_by_class
+  asset.create_subobject:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Create a named subobject OWNED by an existing asset, inside that asset's package, and return its object path for later property writes. This is the missing half of editing a data asset whose payload is named subobjects referenced from a struct array: set_property, bulk_set_properties and append_array_elements can edit the entries, and this makes a new one. Component classes are accepted, unlike create_asset_by_class; Actors are refused because they are spawned into a level. className takes a /Script/Module.Class path, which is how you name a plugin class the Python `unreal` module never exposes. The new object is created RF_Standalone and the package is saved in the same call, so it survives the garbage collection that used to eat a fresh object within one call and is reachable by path from the next one. properties are applied to a throwaway instance first, so a bad path or value creates nothing. outer='asset' (default) gives '<asset>.<name>'; outer='package' puts it beside the asset in the package. Params: assetPath, className, name, properties?, outer? (asset|package), onConflict? (reuse|error), save? (default true) (#975)"
+    options:
+      method: create_subobject
   asset.bulk_upsert_data_assets:
     class_path: ue-mcp.bridge
     group: asset
@@ -541,6 +557,12 @@ tasks:
     description: "Import StringTable entries from CSV. Params: assetPath, filePath (or csvPath)"
     options:
       method: import_stringtable
+  asset.import_stringtable_csv:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Import or refresh a String Table from a CSV that is the canonical source, using Unreal's own String Table CSV importer. The CSV is parsed into a throwaway table and checked against expectedKeys FIRST, so a malformed file or a key set that does not match leaves the asset untouched and comes back with missingKeys/unexpectedKeys instead. replaceExisting=true prunes entries the CSV no longer carries, which is what makes the CSV canonical rather than additive; the prune runs after a successful merge so a bad parse can never empty the table. Returns addedKeys, updatedKeys, removedKeys, entry counts and an explicit persisted/saved plus persistError, so a write that did not reach disk is never reported as a success. A relative csvPath is read against the project directory. Params: assetPath, csvPath (or filePath), expectedKeys? (string[]), requireExactKeys? (default false), replaceExisting? (default false), save? (default true) (#978)"
+    options:
+      method: import_stringtable_csv
   asset.add_input_mapping:
     class_path: ue-mcp.bridge
     group: asset
@@ -586,13 +608,13 @@ tasks:
   asset.set_property:
     class_path: ue-mcp.bridge
     group: asset
-    description: "Set a UPROPERTY on any loaded asset (Material, DataAsset, DataTable, SubsurfaceProfile, etc.) using a dotted path. Blueprint paths resolve to the generated-class CDO so you can author its defaults + Instanced sub-object arrays (#568). Walks nested structs, array elements by index, and instanced subobjects internally - no more read-modify-write copies (e.g. `settings.mean_free_path_distance` on a UMaterial, or `Config.Traits[1].Params.Field` on a config asset #527). Value goes through MCPJsonProperty::SetJsonOnProperty so JSON null clears object refs, structs accept {x,y,z}, arrays/maps round-trip. TMap values take { \"Key\": value } or, for struct keys, [{ key: {...}, value: ... }]; a write that cannot store every entry fails and leaves the old value untouched (#820). Params: assetPath, propertyName (dotted path), value (#420)"
+    description: "Set a UPROPERTY on any loaded asset (Material, DataAsset, DataTable, SubsurfaceProfile, etc.) using a dotted path. Blueprint paths resolve to the generated-class CDO so you can author its defaults + Instanced sub-object arrays (#568). Walks nested structs, array elements by index, and instanced subobjects internally - no more read-modify-write copies (e.g. `settings.mean_free_path_distance` on a UMaterial, or `Config.Traits[1].Params.Field` on a config asset #527). Value goes through MCPJsonProperty::SetJsonOnProperty so JSON null clears object refs, structs accept {x,y,z}, arrays/maps round-trip. TMap values take { \"Key\": value } or, for struct keys, [{ key: {...}, value: ... }]; a write that cannot store every entry fails and leaves the old value untouched (#820). The write is saved to the package by default and the result reports persisted plus the package name; a write that could not reach disk (save=false, a protected mount, a read-only file, a refused save) comes back with persisted=false and persistError naming the reason instead of a bare success that reverts on the next editor start (#931). Params: assetPath, propertyName (dotted path), value, save? (default true) (#420)"
     options:
       method: set_asset_property
   asset.append_array_elements:
     class_path: ue-mcp.bridge
     group: asset
-    description: "Append one or more JSON values to a reflected TArray without replacing existing entries. Supports dotted property paths plus native and user-defined USTRUCT elements. All elements are validated before mutation; returns appended indices and rollback data, and leaves the package dirty without saving. Params: assetPath, propertyName, elements"
+    description: "Append one or more JSON values to a reflected TArray without replacing existing entries. Supports dotted property paths plus native and user-defined USTRUCT elements. All elements are validated before mutation; returns appended indices and rollback data. The append is saved to the package by default and reports persisted / persistError the same way set_property does (#931). Params: assetPath, propertyName, elements, save? (default true)"
     options:
       method: append_asset_array_elements
   asset.bulk_set_properties:
@@ -601,6 +623,12 @@ tasks:
     description: "Set dotted UPROPERTY paths on as many as 500 assets in one preflighted batch. Every asset, path, and value is validated before anything is mutated, and every submitted item comes back with its own ok/status/error, so a bad path in item 300 never hides the other 499 verdicts. Default is all-or-nothing: any preflight rejection aborts before a single UObject is touched. Pass continueOnError to apply the items that did pass and keep the rejects reported alongside them. Returns per-property readback, aggregate counts, targeted save results, and a replayable rollback payload covering only the writes that landed. Params: items ([{assetPath, properties}]), save? (default true), dryRun? (default false), continueOnError? (default false)"
     options:
       method: bulk_set_asset_properties
+  asset.bulk_read_properties:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Read the SAME properties off many assets in one call, filtered and aggregated in the editor. read_properties answers one asset per call, which turns a library-wide question (concurrency settings across 582 sound assets, cull distances across 201 foliage types) into a loop. Select with assetPaths[] or directory + classNames[]; propertyNames accepts dotted paths into nested structs (e.g. 'CullDistance.Max'), and a Blueprint path reads its generated-class CDO. Predicates, groupBy and countBy all evaluate here: filtering happens before anything crosses the wire. Absent and null are reported separately, because 'this class has no such property' and 'this property is unset' are different findings; suspect is true for either, and suspectOnly returns just those rows. Never writes, and reports dirtiedPackages. Params: assetPaths? (string[]) OR directory? (+ recursive?, default true), classNames? (string[]), matchSubclasses? (default true), propertyNames (string[], required, max 32), where? ([{field, op, value}] over props.<name>, className, suspect; same operators as level(query_components)), whereMode? (all|any), suspectOnly?, groupBy?, countBy? (string[]), sampleLimit?, countOnly?, limit? (default 200, max 2000), startIndex?, maxAssets? (default 2000, max 20000), outputPath? (write every matched row to a JSON file and return the path instead of the rows) (#909)"
+    options:
+      method: bulk_read_asset_properties
   asset.set_texture_settings_by_type:
     class_path: ue-mcp.bridge
     group: asset
@@ -709,6 +737,24 @@ tasks:
     description: "Inspect StaticMesh collision setup. Params: assetPath. Returns collisionTraceFlag, hasSimple/ComplexCollision, element counts (#177)"
     options:
       method: get_mesh_collision
+  asset.get_mesh_geometry:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Read actual vertex data off a StaticMesh OR SkeletalMesh: per-section positions, uvs, normals and triangle indices, from the engine's render data (no ProceduralMeshComponent plugin needed). Triangle indices are section-local, so they index that section's own positions array; add baseVertexIndex for LOD-global indices. Over 20000 vertices inline is refused - pass dumpToFile to write the full data to a JSON file instead, or narrow with sectionIndex. Params: assetPath, lodIndex? (default 0), sectionIndex? (omit for all sections), include? ([positions|uvs|normals|triangles], omit for all), uvChannel? (default 0), dumpToFile?, outputPath? (#948/#926/#953)"
+    options:
+      method: get_mesh_geometry
+  asset.measure_mesh_geometry:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Measure a StaticMesh OR SkeletalMesh LOD: bounds, dimensions, surfaceArea, volume, triangleCount, vertexCount, isClosed, isManifold, boundaryEdgeCount, nonManifoldEdgeCount. surfaceArea and volume are NAMED fields, never a positional pair (the engine's get_mesh_volume_area returns them in the order opposite to its name, #938). Vertices are welded by position before topology analysis so a UV seam does not read as a hole. Params: assetPath, lodIndex? (default 0), sectionIndex? (omit to measure the whole LOD) (#938/#953)"
+    options:
+      method: measure_mesh_geometry
+  asset.mesh_boolean:
+    class_path: ue-mcp.bridge
+    group: asset
+    description: "Boolean CSG between two StaticMeshes: union, subtract, intersect, trimInside, trimOutside, newPolyGroupInside, newPolyGroupOutside. The target is the mesh being cut and the tool is what cuts it, each placed by its own optional transform. Writes to a SEPARATE output asset by default (outputPath, or '<targetPath>_<Operation>' when omitted); inPlace=true opts into overwriting the target and is the only destructive form. An existing outputPath is refused unless onConflict='replace'. An empty result is refused and nothing is written unless allowEmptyResult=true, because two meshes that never overlap otherwise report success having deleted everything. Returns triangle and vertex counts for both inputs and the result, plus the written asset's triangles, vertices, LOD count, material slots and bounds, so the operation can be verified rather than trusted. dryRun runs the boolean and reports those counts without writing. Materials and simple collision are copied from the target by default; nanite is inherit (match the target) | enable | disable. Needs the Geometry Script engine plugin: without it the call returns reason='geometry_scripting_unavailable' naming what to enable, rather than failing opaquely. Params: operation, targetPath, toolPath, outputPath?, inPlace?, targetTransform?, toolTransform?, lodType? (MaxAvailable|HiResSourceModel|SourceModel|RenderData), lodIndex?, fillHoles? (default true), simplifyOutput? (default true), simplifyPlanarTolerance? (default 0.01), allowEmptyResult?, recomputeNormals?, recomputeTangents?, removeDegenerates?, copyCollisionFromTarget? (default true), copyMaterialsFromTarget? (default true), nanite?, onConflict? (error|replace), dryRun?, save? (default true) (#916)"
+    options:
+      method: mesh_boolean
   asset.migrate:
     class_path: asset.migrate
     group: asset
@@ -800,7 +846,7 @@ tasks:
   asset.diff:
     class_path: ue-mcp.bridge
     group: asset
-    description: "Semantic structural diff between two assets of any type, dispatching on the asset's class. Blueprints are diffed structurally (parent class, variables, functions, components, per-graph node and connection deltas); other asset types report that diffing is not supported yet rather than failing opaquely. Params: assetPath, otherPath"
+    description: "Semantic structural diff between two assets, dispatching on the asset's class. Blueprints: parent class, variables, functions, components, per-graph node and connection deltas. Skeleton and SkeletalMesh: raw bone additions and removals, reparenting (bone, fromParent, toParent), raw-index changes (bone, fromIndex, toIndex), and declared virtual-bone additions and removals, with hierarchyCompatible and editorCompatible reported SEPARATELY. That separation is the point: it answers whether two skeletons are bone-compatible enough to register as Compatible Skeletons or whether a retarget is required, because appending bones (virtual ones especially) leaves the shared hierarchy intact while reparenting an existing bone does not (#879). Deliberately OUT of scope and reported as such: reference-pose transforms (referencePoseCompared=false), export names (exportNamesCompared=false), sockets and retarget sources; structureScope spells the boundary out in the result. Both paths must be the same class. Other asset types report that diffing is not supported yet rather than failing opaquely. Params: assetPath, otherPath"
     options:
       method: diff_asset
 
@@ -808,13 +854,13 @@ tasks:
   blueprint.read:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Read BP structure incl. SCS components AND inherited native components from the CDO (CharacterMesh0, CharMoveComp, etc.). Params: assetPath, includeComponentProperties? (dump UPROPERTY name/type/value per component template; off by default) (#353/#370)"
+    description: "Read BP structure incl. SCS components AND inherited native components from the CDO (CharacterMesh0, CharMoveComp, etc.). assetPath also accepts a World/umap path such as /Game/Maps/SomeLevel, which resolves to that map's level script Blueprint at PersistentLevel.<MapName>; the result reports blueprintPath and isLevelScript so you can see what answered (#942). Params: assetPath, includeComponentProperties? (dump UPROPERTY name/type/value per component template; off by default) (#353/#370)"
     options:
       method: read_blueprint
   blueprint.list_variables:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "List variables with name, type, guid, category, tooltip, exposeOnSpawn, blueprintReadOnly, private, and editFlag ('EditAnywhere'|'EditDefaultsOnly'|'EditInstanceOnly'|'none'). instanceEditable is true only for EditAnywhere and EditInstanceOnly - EditDefaultsOnly is class-defaults-only and reports false (#744). private is reported separately because it is a Blueprint-graph access flag, not a details-panel one. Params: assetPath"
+    description: "List variables with name, type, guid, category, tooltip, exposeOnSpawn, blueprintReadOnly, private, and editFlag ('EditAnywhere'|'EditDefaultsOnly'|'EditInstanceOnly'|'none'). instanceEditable is true only for EditAnywhere and EditInstanceOnly - EditDefaultsOnly is class-defaults-only and reports false (#744). private is reported separately because it is a Blueprint-graph access flag, not a details-panel one. Pass includeValues to add each variable's RESOLVED default from the generated-class CDO (value, valueText, cppType, declaringClass, inherited) plus one packageDirty/persisted verdict for the listing; off by default so the declaration list stays small (#902). Params: assetPath, includeValues?"
     options:
       method: list_blueprint_variables
   blueprint.list_functions:
@@ -826,19 +872,19 @@ tasks:
   blueprint.read_graph:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Read graph nodes. Supports pagination, file dumps, and title/class node filters. With includeDefaults, literal FText pin values are returned inline (defaultValue falls back to the pin's text literal, plus an explicit defaultTextValue) and object-ref pins report defaultObject (#743). Params: assetPath, graphName, offset?, limit?, includePins?, includeDefaults?, includeComments?, dumpToFile?, outputPath?, titleFilter?, classFilter? (#560)"
+    description: "Read graph nodes. Supports pagination, file dumps, and title/class node filters. assetPath accepts a World/umap path, resolved to that map's level script Blueprint (#942). With includeDefaults, literal FText pin values are returned inline (defaultValue falls back to the pin's text literal, plus an explicit defaultTextValue) and object-ref pins report defaultObject (#743). Params: assetPath, graphName, offset?, limit?, includePins?, includeDefaults?, includeComments?, dumpToFile?, outputPath?, titleFilter?, classFilter? (#560)"
     options:
       method: read_blueprint_graph
   blueprint.read_graph_summary:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Lightweight graph summary (nodes+edges only, ~10KB). Filterable node list. Params: assetPath, graphName?, titleFilter?, classFilter? (#560)"
+    description: "Lightweight graph summary (~10KB). Returns nodes plus wires split across two arrays, execEdges and dataEdges, not a single combined edges array. Filterable node list. assetPath accepts a World/umap path, resolved to that map's level script Blueprint (#942). Params: assetPath, graphName?, titleFilter?, classFilter? (#560)"
     options:
       method: read_blueprint_graph_summary
   blueprint.get_execution_flow:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Trace exec pins from an entry point. Params: assetPath, graphName?, entryPoint?"
+    description: "Trace exec pins from an entry point. assetPath accepts a World/umap path, resolved to that map's level script Blueprint (#942). Params: assetPath, graphName?, entryPoint?"
     options:
       method: get_blueprint_execution_flow
   blueprint.get_dependencies:
@@ -979,6 +1025,12 @@ tasks:
     description: "Set default value on a BP variable. Params: assetPath, name, value"
     options:
       method: set_variable_default
+  blueprint.get_variable_default:
+    class_path: ue-mcp.bridge
+    group: blueprint
+    description: "Read the RESOLVED default of a BP variable off the generated-class CDO, which is what a write-compile-readback loop has to compare against (list_variables alone only proves the variable exists). Returns value (typed JSON), valueText (UE export text), cppType, declaringClass/declaringClassPath and inherited. It also states whether that value is on DISK: packageDirty is the package's own unsaved-changes flag and persisted is its inverse, so a value written into the CDO but never saved reports persisted=false with a persistenceNote instead of reading back as if it had landed (#931). authoredDefault/matchesAuthoredDefault appear when the variable carries a non-empty authored default string that the next recompile could apply. Params: assetPath, name (#902)"
+    options:
+      method: get_blueprint_variable_default
   blueprint.compile:
     class_path: ue-mcp.bridge
     group: blueprint
@@ -991,6 +1043,12 @@ tasks:
     description: "List node types. Params: category?, includeFunctions?"
     options:
       method: list_node_types
+  blueprint.search_call_sites:
+    class_path: ue-mcp.bridge
+    group: blueprint
+    description: "Find authored call-site NODES for named functions across a whole directory in ONE call, including nested and collapsed graphs. Replaces the enumerate-every-Blueprint / read-every-graph sweep that overwhelms the bridge on a large project. Narrowing is done by the Asset Registry BEFORE any package is loaded: candidates come from a recursive Blueprint filter over `directory`, then every candidate whose package does not depend on a package declaring one of `functionNames` is ruled out unloaded. That step is reported (stats.narrowedByRegistry, stats.blueprintsSkippedByRegistry) and can be turned off with narrowByRegistry=false; it is also skipped automatically when a requested name has no resolvable declaring class, so a filter never silently drops a hit. Each hit carries assetPath, packageName, graphName + graphSelector (the same selector list_graphs reports, so it is unambiguous for duplicate nested graph names) + graphObjectPath, nodeId, nodeTitle, memberName, resolvedFunction (absent with unresolvedTarget=true when the function is gone, which is itself an audit finding), declaringClass/declaringClassPath, parentCall, and pinDefaults for unlinked input pins. Pass includeNeighbours for the immediate execution and data neighbours of each call. Level script Blueprints are NOT registry assets (they live inside their map package), so they are excluded unless you pass includeLevelScripts, which sweeps the World assets under the directory and costs a full map load each. Bounded and paginated: limit (default 200, max 1000) + offset, maxBlueprints (default 2000) caps package loads, and dumpToFile writes the whole result set to a JSON file the way read_graph does. Params: functionNames[], className?, directory? (default /Game), includeNestedGraphs? (default true), includeLevelScripts?, includeNeighbours?, narrowByRegistry?, offset?, limit?, maxBlueprints?, dumpToFile?, outputPath? (#945)"
+    options:
+      method: search_blueprint_call_sites
   blueprint.search_node_types:
     class_path: ue-mcp.bridge
     group: blueprint
@@ -1012,7 +1070,7 @@ tasks:
   blueprint.override_function:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Override an inherited interface implementation (inherited via the parent class) or an overridable parent virtual function, with the matching signature so it actually binds as the override (create_function makes a blank, non-binding graph). Returns kind ('function' with a graphName, or 'event' with a nodeId in EventGraph). Event-shaped functions become override events unless preferFunction=true. Pass interfacePath to also implement the interface first if it is not present. Then wire the body with add_node (use nodeClass='CallParent' to chain to the base implementation). Params: assetPath, functionName, source? ('auto'|'interface'|'parent', advisory), preferFunction?, interfacePath? (#688)"
+    description: "Override an inherited interface implementation (inherited via the parent class) or an overridable parent virtual function, with the matching signature so it actually binds as the override (create_function makes a blank, non-binding graph). Returns kind ('function' with a graphName, or 'event' with a nodeId in EventGraph). Event-shaped functions become override events unless preferFunction=true. This is the path for BehaviorTree TASK Blueprint entry points: ReceiveExecuteAI and ReceiveAbortAI on a UBTTask_BlueprintBase child come back as kind='event' in EventGraph, and their FinishExecute / FinishAbort terminators are placed with add_node nodeClass='CallFunction', className='/Script/AIModule.BTTask_BlueprintBase' (#886). Pass interfacePath to also implement the interface first if it is not present. Then wire the body with add_node (use nodeClass='CallParent' to chain to the base implementation). An animation layer override in a child Anim BP is created as an AnimationGraph on the animation schema, mirrored from the graph that declared the layer, so it actually binds instead of compiling to an inert K2 function graph; the result reports graphClass, schemaClass and animationGraph (#894). Params: assetPath, functionName, source? ('auto'|'interface'|'parent', advisory), preferFunction?, interfacePath? (#688)"
     options:
       method: override_function
   blueprint.list_overridable_functions:
@@ -1024,13 +1082,13 @@ tasks:
   blueprint.list_graphs:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "List all graphs in a blueprint. Returns selector/objectPath plus duplicateIndex/duplicateCount; pass selector back as graphName for unambiguous nested graph edits. Params: assetPath"
+    description: "List all graphs in a blueprint. Returns selector/objectPath plus duplicateIndex/duplicateCount; pass selector back as graphName for unambiguous nested graph edits. assetPath accepts a World/umap path, resolved to that map's level script Blueprint (#942). Params: assetPath"
     options:
       method: list_blueprint_graphs
   blueprint.resolve_graph:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Resolve a Blueprint graph name to selectors accepted by read_graph/add_node/connect_pins and other graph actions. Duplicate nested AnimBP names such as Locomotion return every match with selectors like Locomotion[3], object paths, and ambiguity metadata. Params: assetPath, graphName"
+    description: "Resolve a Blueprint graph name to selectors accepted by read_graph/add_node/connect_pins and other graph actions. assetPath accepts a World/umap path, resolved to that map's level script Blueprint (#942). Duplicate nested AnimBP names such as Locomotion return every match with selectors like Locomotion[3], object paths, and ambiguity metadata. Params: assetPath, graphName"
     options:
       method: resolve_blueprint_graph
   blueprint.add_event_dispatcher:
@@ -1069,6 +1127,12 @@ tasks:
     description: "Dump ALL UPROPERTYs on a BP component template incl. array contents (#105). Params: assetPath, componentName"
     options:
       method: read_component_properties
+  blueprint.get_component_collision:
+    class_path: ue-mcp.bridge
+    group: blueprint
+    description: "Read a component's EFFECTIVE collision: enabled mode, object type, profile name, and the resolved Block/Overlap/Ignore per channel. read_component_properties cannot answer this - it exports BodyInstance as text, and that text carries only CollisionProfileName plus the responses that DIFFER from the profile, so a response INHERITED from the profile (the Pawn profile blocking Camera) is invisible in it. Each channel reports response (what GetCollisionResponseToChannel actually returns), profileResponse, and overridesProfile, so you can see whether an answer came from the profile or from the component. assetPath takes a Blueprint path OR a native class (/Script/MyGame.MyCharacter or 'MyCharacter'), which is how you verify a C++ constructor's SetCollisionResponseToChannel really landed in the CDO after a rebuild. Project trace and object channels appear under their configured names with enumName alongside. channel narrows it to one; includeAllChannels adds the unused GameTraceChannel slots. Params: assetPath, componentName, channel?, includeAllChannels? (#925)"
+    options:
+      method: get_component_collision
   blueprint.read_node_property:
     class_path: ue-mcp.bridge
     group: blueprint
@@ -1078,13 +1142,13 @@ tasks:
   blueprint.reparent_component:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Reparent an SCS component under a new parent (#115). Params: assetPath, componentName, newParent"
+    description: "Reparent an SCS component under a new parent, recompile and save (#115). Refuses before touching anything when the Blueprint's .uasset is read-only on disk (not checked out) or on a protected mount, and reports saved plus saveError when the write did not reach disk (#932). Params: assetPath, componentName, newParent"
     options:
       method: reparent_component
   blueprint.reparent:
     class_path: ue-mcp.bridge
     group: blueprint
-    description: "Change a Blueprint's ParentClass and recompile (#138). Params: assetPath, parentClass (short name or full path)."
+    description: "Change a Blueprint's ParentClass, recompile and save (#138). The save is part of the operation, so writability is checked FIRST: a read-only .uasset (never checked out of source control) or a protected mount returns reason='package_not_writable' naming the file, with nothing reparented, instead of taking the editor down with a fatal save error (#932). A write that still did not reach disk comes back with saved=false and saveError. Params: assetPath, parentClass (short name or full path)."
     options:
       method: reparent_blueprint
   blueprint.flush_ich:
@@ -1093,6 +1157,12 @@ tasks:
     description: "Flush orphaned InheritableComponentHandler override records (invalid component-override entries invisible to read/remove_component). Recompiles + saves. Params: assetPath. Returns recordsBefore/After/Removed (#580)"
     options:
       method: flush_inheritable_component_handler
+  blueprint.flush_component_templates:
+    class_path: ue-mcp.bridge
+    group: blueprint
+    description: "Flush orphaned UBlueprint::ComponentTemplates left by deleted Add Component nodes using Unreal's native maintenance routine. Live Add Component templates are preserved. Recompiles + saves only when changed. Params: assetPath. Returns recordsBefore/After/Removed and removed template identities."
+    options:
+      method: flush_blueprint_component_templates
   blueprint.set_actor_tick_settings:
     class_path: ue-mcp.bridge
     group: blueprint
@@ -1168,7 +1238,7 @@ tasks:
   level.get_outliner:
     class_path: ue-mcp.bridge
     group: level
-    description: "List actors (each row includes editorHidden - hidden in the viewport but still rendering in game). Params: classFilter?, nameFilter?, editorHidden? (filter to only hidden/only visible), world? (editor|pie|auto), limit?, includeStreaming? (#717)"
+    description: "List actors (each row includes editorHidden - hidden in the viewport but still rendering in game, and folderPath). classFilter is a case-sensitive SUBSTRING over the class name by default; pass exactClass=true to require the exact class instead. folderPath matches one World Outliner folder exactly and folderPathPrefix matches that folder and everything nested under it, so narrowing a folder to a single class no longer needs a get_actor_details round trip per entry (#911). Params: classFilter?, exactClass?, nameFilter? (case-sensitive substring over the internal name OR the label), folderPath?, folderPathPrefix?, editorHidden? (filter to only hidden/only visible), world? (editor|pie|auto), limit?, includeStreaming? (#717)"
     options:
       method: get_world_outliner
   level.set_editor_visibility:
@@ -1186,25 +1256,25 @@ tasks:
   level.delete_actor:
     class_path: ue-mcp.bridge
     group: level
-    description: "Remove actor. Params: actorLabel"
+    description: "Remove actor. Params: actorLabel OR actorPath. A label matching several actors is refused with the candidates rather than deleting one of them (#983)"
     options:
       method: delete_actor
   level.get_actor_details:
     class_path: ue-mcp.bridge
     group: level
-    description: "Inspect actor. Params: actorLabel OR actorPath, includeProperties?, propertyName?, world? (editor|pie)"
+    description: "Inspect actor. With includeProperties=true, a UPROPERTY declared as a C-style FIXED ARRAY (int32 Foo[3], one FProperty with ArrayDim 3) comes back as a JSON ARRAY of its elements plus arrayDim, instead of element 0 presented as the whole value. That mattered on RecastNavMesh NavMeshResolutionParams, whose three entries are the Low, Default and High generation tiers, and where reporting only Low sent a user tuning cell sizes against numbers the engine was not using (#927). Params: actorLabel OR actorPath, includeProperties?, propertyName?, world? (editor|pie)"
     options:
       method: get_actor_details
   level.move_actor:
     class_path: ue-mcp.bridge
     group: level
-    description: "Transform actor. Pass world:pie to move a live PIE actor (resolves labels/names from get_outliner {world:pie}) (#586). Params: actorLabel, location?, rotation?, scale?, world? (editor|pie)"
+    description: "Transform actor. Pass world:pie to move a live PIE actor (resolves labels/names from get_outliner {world:pie}) (#586). Params: actorLabel OR actorPath, location?, rotation?, scale?, world? (editor|pie)"
     options:
       method: move_actor
   level.aim_actor_at:
     class_path: ue-mcp.bridge
     group: level
-    description: "Rotate an actor so its forward (+X) points at a target. Params: actorLabel, targetPoint (Vec3) OR targetActor (label), roll? (default 0), world? (editor|pie) (#566)"
+    description: "Rotate an actor so its forward (+X) points at a target. Params: actorLabel OR actorPath, targetPoint (Vec3) OR targetActor (label) OR targetActorPath, roll? (default 0), world? (editor|pie) (#566/#983)"
     options:
       method: aim_actor_at
   level.nav_project_point:
@@ -1216,7 +1286,7 @@ tasks:
   level.select:
     class_path: ue-mcp.bridge
     group: level
-    description: "Select actors. Params: actorLabels[]"
+    description: "Select actors. A label naming several actors selects all of them; selectedPaths reports exactly which. Params: actorLabels[] and/or actorPaths[] (#983)"
     options:
       method: select_actors
   level.get_selected:
@@ -1228,25 +1298,31 @@ tasks:
   level.add_component:
     class_path: ue-mcp.bridge
     group: level
-    description: "Add component to actor. Params: actorLabel, componentClass, componentName?"
+    description: "Add component to actor. Params: actorLabel OR actorPath, componentClass, componentName?"
     options:
       method: add_component_to_actor
   level.remove_component:
     class_path: ue-mcp.bridge
     group: level
-    description: "Remove instance component from a level actor by name. Idempotent: returns alreadyDeleted=true if no matching component exists. Params: actorLabel, componentName (#426)"
+    description: "Remove instance component from a level actor by name. Idempotent: returns alreadyDeleted=true if no matching component exists. Params: actorLabel OR actorPath, componentName (#426)"
     options:
       method: remove_component_from_actor
   level.set_component_property:
     class_path: ue-mcp.bridge
     group: level
-    description: "Set component prop. Pass value=null to clear a TObjectPtr/SoftObject/WeakObject/UClass/Interface reference (#420). Resolves inherited/SCS components on placed Blueprint instances case-insensitively, and refreshes the scene transform after RelativeLocation/RelativeRotation/RelativeScale3D writes (#539). Pass world='pie' (with optional pieInstance) to write to a LIVE PIE component rather than the editor one (#763). Params: actorLabel, componentName, propertyName, value, world?, pieInstance?"
+    description: "Set component prop. A C-style FIXED ARRAY element is addressed by index in the dotted path ('Foo[2].Bar'); without an index every write lands on element 0 (#927). Pass value=null to clear a TObjectPtr/SoftObject/WeakObject/UClass/Interface reference (#420). Resolves inherited/SCS components on placed Blueprint instances case-insensitively, and refreshes the scene transform after RelativeLocation/RelativeRotation/RelativeScale3D writes (#539). Pass world='pie' (with optional pieInstance) to write to a LIVE PIE component rather than the editor one (#763). Params: actorLabel OR actorPath, componentName, propertyName, value, world?, pieInstance?"
     options:
       method: set_component_property
+  level.nudge_component:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Make a small spatial adjustment to an exact named SceneComponent without editing raw relative Euler angles. For visual tuning, normalize the user's request as target + reference frame + viewpoint + operation + amount + constraints. Character left/right means actor-relative; viewer-relative directions must be stated explicitly; clockwise/counterclockwise requires a viewpoint. Infer obvious context and echo the normalized interpretation before applying; ask one concise question only when the frame or viewpoint cannot be resolved. Translation uses forward/right/up centimetres in frame=world|actor|parent|component. Rotation is one axis-angle quaternion delta in that frame, so attached/socketed components avoid gimbal-like pitch/roll drift. Unspecified location, rotation and scale are preserved. After a visual nudge, verify with a fixed-pose close packet: for character body attachments, stop actor/root movement and freeze the parent skeletal pose before the baseline, then hold both across every probe and capture; light the contact area from the front, capture front/side/top/three-quarter views with the target filling roughly 50-75% of the frame, and reject dark or wide-only evidence. Returns before/after relative and world transforms, resolved world deltas, and exact rollback data. Params: actorLabel OR actorPath, componentName, frame? (default actor), translationDelta? ({forwardCm?,rightCm?,upCm?}), axisRotation? ({axis:forward|right|up,degrees}), scaleMultiplier? (>0), world? (editor|pie), pieInstance?"
+    options:
+      method: nudge_component
   level.get_component_details:
     class_path: ue-mcp.bridge
     group: level
-    description: "Read a placed actor's component transforms. With componentName returns that component's relative+world location/rotation/scale, class, and attach parent; without it lists every component. With includeValues=true also dumps arbitrary UPROPERTY values (custom fields, CharacterMovement MaxWalkSpeed, etc.); world='pie' reads the live PIE instance (#539/#584). Params: actorLabel, componentName?, includeValues?, propertyNames? (filter), world? (editor|pie)"
+    description: "Read a placed actor's component transforms. With componentName returns that component's relative+world location/rotation/scale, class, and attach parent; without it lists every component. With includeValues=true also dumps arbitrary UPROPERTY values (custom fields, CharacterMovement MaxWalkSpeed, etc.); world='pie' reads the live PIE instance (#539/#584). Params: actorLabel OR actorPath, componentName?, includeValues?, propertyNames? (filter), world? (editor|pie)"
     options:
       method: get_component_details
   level.get_current:
@@ -1261,12 +1337,18 @@ tasks:
     description: "Load level. Params: levelPath"
     options:
       method: load_level
+  level.clear_level_script:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Preview or remove every node and member variable from the currently loaded persistent level's Level Blueprint. Defaults to dryRun=true. Actors are untouched; save=true saves only the current level after a successful compile. Params: dryRun?, save?"
+    options:
+      method: clear_level_script
   level.save:
     class_path: ue-mcp.bridge
     group: level
-    description: Save current level
+    description: "Save the level currently being edited, and say what happened to each package. Takes the same package path editor(save_dirty) uses, so the two can no longer disagree about whether one package was written: the old action returned a bare {success:false, error:'Failed to save current level'} on a map that save_dirty then wrote seconds later, and a false failure makes an agent redo work that was already on disk (#964). On a World Partition or one-file-per-actor map the actors live in their own packages, so those are saved too and each is reported separately. A package that was already clean is reported as skipped, not as a failure. On failure the error names the package, the resolved file, whether that file exists and is read-only, the engine's ESavePackageResult and the error/warning lines the save itself emitted. Saving during PIE is refused by name rather than failing silently. Params: force? (write even a clean package, default false), includeExternalActors? (default true)"
     options:
-      method: save_current_level
+      method: save_level
   level.list:
     class_path: ue-mcp.bridge
     group: level
@@ -1294,7 +1376,7 @@ tasks:
   level.set_volume_properties:
     class_path: ue-mcp.bridge
     group: level
-    description: "Edit volume. Params: actorLabel, properties"
+    description: "Edit volume. Params: actorLabel OR actorPath, properties"
     options:
       method: set_volume_properties
   level.spawn_light:
@@ -1306,13 +1388,13 @@ tasks:
   level.set_light_properties:
     class_path: ue-mcp.bridge
     group: level
-    description: "Edit a light OR SkyLight (intensity/color now work on SkyLight too, #608). Params: actorLabel, intensity?, color?, rotation? (DirectionalLight sun angle), mobility? (static|stationary|movable), recaptureSky?, volumetricScatteringIntensity?, sourceRadius? (point/spot), innerConeAngle?/outerConeAngle? (spot)"
+    description: "Edit a light OR SkyLight (intensity/color now work on SkyLight too, #608). Params: actorLabel OR actorPath, intensity?, color?, rotation? (DirectionalLight sun angle), mobility? (static|stationary|movable), recaptureSky?, volumetricScatteringIntensity?, sourceRadius? (point/spot), innerConeAngle?/outerConeAngle? (spot)"
     options:
       method: set_light_properties
   level.set_fog_properties:
     class_path: ue-mcp.bridge
     group: level
-    description: "Edit ExponentialHeightFog incl. volumetric fog (#608). Params: actorLabel?, fogDensity?, fogHeightFalloff?, startDistance?, fogInscatteringColor?, enableVolumetricFog?, volumetricFogScatteringDistribution?, volumetricFogExtinctionScale?, volumetricFogDistance?, volumetricFogAlbedo?"
+    description: "Edit ExponentialHeightFog incl. volumetric fog (#608). Params: actorLabel? OR actorPath?, fogDensity?, fogHeightFalloff?, startDistance?, fogInscatteringColor?, enableVolumetricFog?, volumetricFogScatteringDistribution?, volumetricFogExtinctionScale?, volumetricFogDistance?, volumetricFogAlbedo?"
     options:
       method: set_fog_properties
   level.get_actors_by_class:
@@ -1324,9 +1406,15 @@ tasks:
   level.get_actors_by_component_class:
     class_path: ue-mcp.bridge
     group: level
-    description: "List actors that own a component of a given class (exact or substring match). Returns each actor plus its matchedComponents. Params: componentClass, world? (editor|pie) (#582)"
+    description: "List actors that own a component of a given class (exact or substring match). Returns each actor plus its matchedComponents. For a whole-map question use level(query_components), which filters and counts in the editor instead of returning every candidate. Params: componentClass, world? (editor|pie) (#582)"
     options:
       method: get_actors_by_component_class
+  level.query_components:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Ask one level-wide question about components and get the answer, not the candidates. Selection, projection, predicates, grouping and counts all evaluate in the editor, so a 4,000 actor map costs one call instead of thousands. Computed (not raw) values: shadow.effectiveCastShadow crosses CastShadow with visibility, bHiddenInGame, owner hidden and bCastHiddenShadow, so a hidden debug marker with CastShadow=true is correctly NOT a shadow caster; every mobility and collision enum comes back as a clean name (Movable, QueryAndPhysics), never as EComponentMobility.MOVABLE; navigation.effectiveNavRelevant crosses bCanEverAffectNavigation with collision and custom navigable geometry. Never writes, and reports dirtiedPackages so you can see that. Field groups for 'fields' and for dot paths: transform, bounds, localBounds (unscaled, includes UBoxComponent BoxExtent), shadow, nanite, navigation, tick, materials (per slot: effective path, componentOverride vs meshDefault, slot name/index, isNull, isWorldGrid), mesh (adds instanceCount + hierarchical for an ISM/HISM, #986), decal, health (roll-up incl. suspect). A dot path used in where/groupBy/countBy turns its own group on. groupBy and countBy COUNT COMPONENTS, not instances: one HISM holding 5,000 instances counts as 1, so for per-mesh placement totals use level(summarize_static_mesh_usage), which sums instances and counts distinct actors. Params: componentClass?, actorClass?, matchSubclasses? (default true), componentNameContains?, actorLabelPrefix?, actorLabelContains?, actorTag?, folderPath?, folderPathPrefix?, fields? (string[]), propertyNames? (string[], projected component UPROPERTYs under props.*), where? ([{field, op, value}]; ops eq/ne/lt/lte/gt/gte/contains/notContains/startsWith/endsWith/in/notIn/exists/notExists/isNull/isNotNull/isTrue/isFalse), whereMode? (all|any, default all), suspectOnly?, groupBy?, countBy? (string[]), sampleLimit?, countOnly?, limit? (default 200, max 2000), startIndex?, duplicateTransformTolerance? (cm; reports duplicateTransformGroups), levelPath? (temporarily opens another map: refuses if anything is dirty and restores the map that was open), world? (editor|pie) (#910/#943/#912)"
+    options:
+      method: query_components
   level.count_actors_by_class:
     class_path: ue-mcp.bridge
     group: level
@@ -1342,7 +1430,7 @@ tasks:
   level.set_water_body_property:
     class_path: ue-mcp.bridge
     group: level
-    description: "Set a property on an actor's WaterBodyComponent (ShapeDilation, WaterLevel, etc.). Params: actorLabel, propertyName, value. Requires Water plugin. For spline edits use level(set_spline_points) (#151)"
+    description: "Set a property on an actor's WaterBodyComponent (ShapeDilation, WaterLevel, etc.). Params: actorLabel OR actorPath, propertyName, value. Requires Water plugin. For spline edits use level(set_spline_points) (#151)"
     options:
       method: set_water_body_property
   level.build_lighting:
@@ -1354,19 +1442,19 @@ tasks:
   level.get_spline_info:
     class_path: ue-mcp.bridge
     group: level
-    description: "Read a spline component's points, closedLoop, length, and per-point tangents/types. Works in editor or PIE (#553). Optional componentName picks a specific (custom) spline; projectPoint (Vec3) returns the closest location, inputKey, distanceAlongSpline, distanceToSpline, and tangent (#555). Params: actorLabel, componentName?, world? (editor|pie), projectPoint?"
+    description: "Read a spline component's points, closedLoop, length, and per-point tangents/types. Works in editor or PIE (#553). Optional componentName picks a specific (custom) spline; projectPoint (Vec3) returns the closest location, inputKey, distanceAlongSpline, distanceToSpline, and tangent (#555). Params: actorLabel OR actorPath, componentName?, world? (editor|pie), projectPoint?"
     options:
       method: get_spline_info
   level.set_spline_points:
     class_path: ue-mcp.bridge
     group: level
-    description: "Set spline points. Params: actorLabel, points[], closedLoop?"
+    description: "Set spline points. Params: actorLabel OR actorPath, points[], closedLoop?"
     options:
       method: set_spline_points
   level.set_actor_material:
     class_path: ue-mcp.bridge
     group: level
-    description: "Set material on actor. Params: actorLabel, materialPath, slotIndex?"
+    description: "Set material on actor. Params: actorLabel OR actorPath, materialPath, slotIndex?"
     options:
       method: set_actor_material
   level.get_world_settings:
@@ -1390,13 +1478,13 @@ tasks:
   level.get_component_tree:
     class_path: ue-mcp.bridge
     group: level
-    description: "Deep component-tree dump for an actor. Returns per-component: name, class, attachParent, attachSocket, mobility, visibility, relative+world transforms, tags. For PrimitiveComponents adds collisionProfile/collisionEnabled/bounds/castShadow. For StaticMeshComponent adds staticMesh + materials[]. For SkeletalMeshComponent adds skeletalMesh + skeleton + materials[]. For NiagaraComponent adds niagara{asset,active,visible} and for AudioComponent audio{sound,playing} - runtime FX state in PIE (#581). Params: actorLabel | actorPath, world? (editor|pie), componentClass? (substring filter), includeProperties? (dump UPROPERTY name/type/value per component) (#240/#241/#302/#320/#370/#353/#581)"
+    description: "Deep component-tree dump for an actor. Returns per-component: name, class, attachParent, attachSocket, mobility, visibility, relative+world transforms, tags. For PrimitiveComponents adds collisionProfile/collisionEnabled/bounds/castShadow. For StaticMeshComponent adds staticMesh + materials[], and for an ISM/HISM also instanceCount plus instanced{hierarchical,numCustomDataFloats} - the number you need before deciding whether to touch an ISM at all, since a component holding three instances and one holding three hundred thousand were otherwise the same row (#986). For SkeletalMeshComponent adds skeletalMesh + skeleton + materials[]. For NiagaraComponent adds niagara{asset,active,visible} and for AudioComponent audio{sound,playing} - runtime FX state in PIE (#581). Params: actorLabel OR actorPath, world? (editor|pie), componentClass? (substring filter), includeProperties? (dump UPROPERTY name/type/value per component) (#240/#241/#302/#320/#370/#353/#581)"
     options:
       method: get_component_tree
   level.get_relative_transform:
     class_path: ue-mcp.bridge
     group: level
-    description: "Compute target's transform in reference's local space (location/rotation/scale). Common dungeon/calibration workflow. Params: target (actor label), reference (actor label), world? (#386/#387)"
+    description: "Compute target's transform in reference's local space (location/rotation/scale). Common dungeon/calibration workflow. Params: target/targetLabel (actor label) OR targetPath, reference/referenceLabel OR referencePath, world? (#386/#387/#983)"
     options:
       method: get_relative_transform
   level.resolve_actor:
@@ -1408,37 +1496,37 @@ tasks:
   level.set_actor_property:
     class_path: ue-mcp.bridge
     group: level
-    description: "Set per-instance UPROPERTY on a level actor. Params: actorLabel ('WorldSettings' targets the world settings actor), propertyName (dotted paths like 'Foo.Bar' supported), value (string/number/bool/object/array; a label string resolves to an AActor* ref, and a JSON array of labels populates a TArray of actor refs #538), force? (bypass EditDefaultsOnly), world? (editor|pie) (#202/#230)"
+    description: "Set per-instance UPROPERTY on a level actor. Params: actorLabel OR actorPath ('WorldSettings' as the label targets the world settings actor), propertyName (dotted paths like 'Foo.Bar' supported, and a C-style FIXED ARRAY element is addressed by index: 'NavMeshResolutionParams[1].CellSize' reaches the Default tier rather than always writing element 0 (#927)), value (string/number/bool/object/array; a label string resolves to an AActor* ref, and a JSON array of labels populates a TArray of actor refs #538), force? (bypass EditDefaultsOnly), world? (editor|pie) (#202/#230)"
     options:
       method: set_actor_property
   level.read_actor_motion:
     class_path: ue-mcp.bridge
     group: level
-    description: "Snapshot motion telemetry for one or many actors: location, rotation, velocity, scale, angularVelocity (when simulating physics), grounded + distanceToGround (downward 200u trace). Defaults to the PIE world with editor fallback. Loop at your sample interval for long telemetry probes. Params: actorLabel? OR actorLabels (string[]), world? ('pie'|'editor'), pieInstance? (#453)"
+    description: "Snapshot motion telemetry for one or many actors: location, rotation, velocity, scale, angularVelocity (when simulating physics), grounded + distanceToGround (downward 200u trace). Defaults to the PIE world with editor fallback. Loop at your sample interval for long telemetry probes. Params: actorLabel? OR actorLabels (string[]) OR actorPath? OR actorPaths (string[]), world? ('pie'|'editor'), pieInstance? (#453)"
     options:
       method: read_actor_motion
   level.add_hismc_instances:
     class_path: ue-mcp.bridge
     group: level
-    description: "Bulk-add transforms to a HISMC / ISMC component on an actor (Python add_instance crashes on UE 5.7; this is the C++ path). Params: actorLabel, componentName? (default: first ISMC/HISMC found), transforms ([{location: {x,y,z}, rotation?: {pitch,yaw,roll}, scale?: {x,y,z}}]), worldSpace? (default true) (#434)"
+    description: "Bulk-add transforms to a HISMC / ISMC component on an actor (Python add_instance crashes on UE 5.7; this is the C++ path). Params: actorLabel OR actorPath, componentName? (default: first ISMC/HISMC found), transforms ([{location: {x,y,z}, rotation?: {pitch,yaw,roll}, scale?: {x,y,z}}]), worldSpace? (default true) (#434)"
     options:
       method: add_hismc_instances
   level.get_instance_transforms:
     class_path: ue-mcp.bridge
     group: level
-    description: "Read every instance transform on an actor's ISMC/HISMC. Params: actorLabel, componentName? (default first), worldSpace? (default true). Returns instances[{index, location, rotation, scale}] (#697)"
+    description: "Read every instance transform on an actor's ISMC/HISMC. Params: actorLabel OR actorPath, componentName? (default first), worldSpace? (default true). Returns instances[{index, location, rotation, scale}] (#697)"
     options:
       method: get_instance_transforms
   level.update_instance_transform:
     class_path: ue-mcp.bridge
     group: level
-    description: "Update a single ISMC/HISMC instance by index; unspecified location/rotation/scale are preserved. Params: actorLabel, componentName?, index, location?, rotation?, scale?, worldSpace? (default true) (#697)"
+    description: "Update a single ISMC/HISMC instance by index; unspecified location/rotation/scale are preserved. Params: actorLabel OR actorPath, componentName?, index, location?, rotation?, scale?, worldSpace? (default true) (#697)"
     options:
       method: update_instance_transform
   level.remove_instance:
     class_path: ue-mcp.bridge
     group: level
-    description: "Remove a single ISMC/HISMC instance by index. Params: actorLabel, componentName?, index (#697)"
+    description: "Remove a single ISMC/HISMC instance by index. Params: actorLabel OR actorPath, componentName?, index (#697)"
     options:
       method: remove_instance
   level.set_nanite_settings:
@@ -1456,25 +1544,43 @@ tasks:
   level.add_post_process_blendable:
     class_path: ue-mcp.bridge
     group: level
-    description: "Add a material blendable (post-process material or MID, e.g. a toon/outline pass) to a PostProcessVolume's WeightedBlendables. Params: actorLabel (the PPV), materialPath, weight? (default 1) (#666)"
+    description: "Add a material blendable (post-process material or MID, e.g. a toon/outline pass) to a PostProcessVolume's WeightedBlendables. Params: actorLabel OR actorPath (the PPV), materialPath, weight? (default 1) (#666)"
     options:
       method: add_post_process_blendable
+  level.set_post_process_settings:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Write post-process settings AND enable the bOverride_<Name> flag for every key given. Inside FPostProcessSettings each value is gated by its own override bit, so writing AutoExposureMinBrightness through the generic property setter stores the number, leaves the bit off, and the engine keeps using the default while the details panel shows the value you asked for. Nothing errors. That is the trap this closes: the auto-enable IS the action. Works off the struct's reflected fields, so every setting is reachable, not just exposure, and a key that has no override bit (WeightedBlendables) says so instead of implying one was enabled. Every key is resolved BEFORE anything is written, so a typo fails the call rather than half-applying it, and each key reports previousValue, newValue, overrideWasEnabled and overrideEnabled read back from the struct. Finds the settings on the actor or on one of its components; a camera with several camera components needs componentName (or propertyName) and is refused by listing the candidates rather than by picking one. The level is left dirty and is NOT saved. Params: actorLabel OR actorPath, settings (object of settingName -> value), enableOverrides? (default true), componentName?, propertyName? (#950)"
+    options:
+      method: set_post_process_settings
+  level.get_post_process_settings:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Read post-process settings with the override bit that decides whether each one applies. onlyOverridden answers 'is this volume actually overriding exposure' directly, instead of dumping the whole FPostProcessSettings as one ExportText blob to parse. Each row is name, type, value, overridden (the bOverride_<Name> bit) and hasOverrideFlag; the bits themselves are not listed as settings of their own, because a caller who writes the bit without the value has changed nothing. Also returns totalSettings and overriddenCount, so 'nothing is overridden here' is a fact rather than an empty list. Params: actorLabel OR actorPath, onlyOverridden? (default false), names? (string[] exact names), nameContains? (substring filter), componentName?, propertyName? (#950)"
+    options:
+      method: get_post_process_settings
+  level.set_fixed_exposure:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Pin exposure on a post-process volume or camera in one call. Writes the same value to AutoExposureMinBrightness and AutoExposureMaxBrightness, which is how the engine disables eye adaptation, and enables both override flags so the values actually apply. Params: actorLabel OR actorPath, exposure (the fixed adaptation brightness), bias? (AutoExposureBias, also override-enabled), componentName?, propertyName? (#950)"
+    options:
+      method: set_fixed_exposure
   level.export_actor_fbx:
     class_path: ue-mcp.bridge
     group: level
-    description: "Export an actor's skeletal/static mesh to FBX and write a metadata sidecar JSON (actor transform, mesh, materials, skeleton) for a downstream bridge (e.g. MetaTailor). Params: actorLabel, outputPath (.fbx). Returns fbx + metadata paths (#637)"
+    description: "Export an actor's skeletal/static mesh to FBX and write a metadata sidecar JSON (actor transform, mesh, materials, skeleton) for a downstream bridge (e.g. MetaTailor). Params: actorLabel OR actorPath, outputPath (.fbx). Returns fbx + metadata paths (#637)"
     options:
       method: export_actor_fbx
   level.spawn_skeletal_mesh_actor:
     class_path: ue-mcp.bridge
     group: level
-    description: "Spawn a SkeletalMeshActor for visual/deform verification. Optional per-slot materials and single-node animation preview (animSequence + loop). Returns actorLabel + boxExtent. The animSequence is written to the component's editable AnimationData, so it persists across a level save/reload instead of only driving the runtime preview and coming back in A-pose (#790). Params: skeletalMesh, label?, location?, rotation?, scale?, materials? (path[]), animSequence?, loop? (#679/#677)"
+    description: "Spawn a SkeletalMeshActor for visual/deform verification. Optional per-slot COMPONENT material overrides (they write the component's OverrideMaterials, not the mesh ASSET's slots, which material(build_material) writes) and single-node animation preview (animSequence + loop). Each entry in materials[] comes back with ok, and a path that does not load or a slot index past the mesh's slot count is reported rather than silently dropped, because a silently dropped assignment reads as one that worked. To dress an actor that is already placed, or to put one material on every slot, use level(set_component_materials) (#946). Returns actorLabel + boxExtent + materialSlotCount. The animSequence is written to the component's editable AnimationData, so it persists across a level save/reload instead of only driving the runtime preview and coming back in A-pose (#790). Params: skeletalMesh, label?, location?, rotation?, scale?, materials? (path[]), animSequence?, loop? (#679/#677)"
     options:
       method: spawn_skeletal_mesh_actor
   level.delete_actors:
     class_path: ue-mcp.bridge
     group: level
-    description: "Bulk-delete actors. Params: at least one of labelPrefix, className, tag; dryRun? to preview. Returns matched/deleted/labels (#220)"
+    description: "Bulk-delete actors. The filter names say exactly what they match, because two level actions taking similar-sounding filters with different semantics is how a caller ends up trusting a wrong zero (#963): labelPrefix is a CASE-SENSITIVE PREFIX over the EDITOR LABEL, labelContains a case-insensitive SUBSTRING over the label, nameContains a case-insensitive substring over the INTERNAL name, className a case-sensitive substring over the class name. get_outliner's nameFilter is looser than any of them (label OR internal name, substring), so a string that selects a set there can select a different set here. When nothing matches, the response explains it rather than returning a bare zero: zeroMatchHint reports how many actors the same string WOULD have matched under the other semantics with samples, and a World Partition map says that only loaded actors were enumerated. classPathContains is a case-insensitive substring of the generated UClass path (e.g. /SurvivalGameKitV2/), so a whole vendor folder can be selected without listing every Blueprint class, and classPathContainsAny is the same match against any of several substrings (#924). Params: at least one of labelPrefix, labelContains, nameContains, className, tag, classPathContains, classPathContainsAny; dryRun? to preview. Returns matched/deleted/labels/classPaths (#220/#924/#963)"
     options:
       method: delete_actors
   level.delete_exact_labeled_actors_in_levels:
@@ -1495,6 +1601,30 @@ tasks:
     description: "World Partition ONLY: pin matched actors so they become resident and the normal actor actions can reach them, or unpin to release. Pinning is used rather than transient streaming because a transient load is dropped again by the streaming system mid-measurement. Takes the same filters as list_actor_descs and refuses to act on more than maxActors (default 256) so an unfiltered call cannot pin a whole map by accident. Reports residentAfter, read back from the world rather than assumed. Needs UE 5.5 or newer. Params: mode? (pin|unpin, default pin), filter?, className?, guids?, bounds?, maxActors?, dryRun? (#746)"
     options:
       method: load_actor_descs
+  level.get_world_partition_settings:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "World Partition ONLY: the streaming knobs on the open map, plus the runtime cell transformer stack. Returns the world partition and runtime hash classes, whether streaming is enabled, the default HLOD layer, and grids[] with each grid's name, CellSize, LoadingRange and a dotted path rooted at the world partition. Grids are found by REFLECTION rather than against a fixed class, so this answers for the RuntimeHashSet layout and the older spatial hash alike. cellTransformers[] lists each stack entry with its class, its instancePath and the current values of the instanced transformer object's editable properties, which no read exposed. Params: none (#985)"
+    options:
+      method: get_world_partition_settings
+  level.set_world_partition_settings:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "World Partition ONLY: write the streaming knobs. cellSize and loadingRange are the shorthand for the primary tuning pair and pick the map's only grid automatically, or name one with grid (its name) or gridPath (the dotted path from get_world_partition_settings). settings takes any dotted path rooted at the world partition, which is how the INSTANCED runtime cell transformer object is configured ('RuntimeCellsTransformerStack[0].Instance.SomeProperty') - that object is read-only through Python and was previously set by hand in the editor UI. Every path is resolved BEFORE anything is written, so a bad path fails the call rather than leaving the map streaming on a mixture nobody asked for, and each write reports previousValue and the value read back. Streaming settings only apply to cells generated after the change, so rebuild the streaming data before measuring. The map package is left dirty and is NOT saved. Params: cellSize?, loadingRange?, grid? (name), gridPath?, settings? (object of dotted path -> value) (#985)"
+    options:
+      method: set_world_partition_settings
+  level.add_runtime_cell_transformer:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "World Partition ONLY: append a WorldPartitionRuntimeCellTransformer to RuntimeCellsTransformerStack and configure its instanced object in the same call. Adding the entry without being able to write the instance would leave the manual editor step exactly where it was, so the instance is created here and properties are applied to it. skipIfPresent (default true) makes a rerun a no-op that reports the existing index instead of stacking duplicates. position inserts at an index rather than appending, because the stack is ordered. Params: transformerClass (short name, Module.Class or /Script path), properties? (object applied to the instance), position? (default -1 = append), skipIfPresent? (default true) (#985)"
+    options:
+      method: add_runtime_cell_transformer
+  level.set_actor_hlod_layer:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Assign (or clear) the per-actor HLODLayer override across every actor a selector picks. Assigning one across 295 InstancedFoliageActors was a Python loop. Goes through the engine's own AActor::SetHLODLayer rather than writing the private property directly, so the result behaves like an edit made in the details panel. hlodLayer=null clears the override. An actor already carrying the requested layer is reported as unchanged rather than counted as an update, so a rerun is readable. Optionally sets bEnableAutoLODGeneration in the same pass, since the layer only matters when the actor is allowed to build HLODs at all. On a World Partition map this dirties one package per actor and does NOT save. Params: hlodLayer (asset path or null), at least one of actorLabels[]/labelPrefix/labelContains/tag/classFilter/folderPath/folderPathPrefix, matchSubclasses? (default true), enableAutoLODGeneration?, dryRun?, transactionLabel? (#985)"
+    options:
+      method: set_actor_hlod_layer
   level.set_actor_folder_path:
     class_path: ue-mcp.bridge
     group: level
@@ -1504,55 +1634,55 @@ tasks:
   level.add_actor_tag:
     class_path: ue-mcp.bridge
     group: level
-    description: "Append a tag to an actor's Tags array. Params: actorLabel, tag (#219)"
+    description: "Append a tag to an actor's Tags array. Params: actorLabel OR actorPath, tag (#219)"
     options:
       method: add_actor_tag
   level.remove_actor_tag:
     class_path: ue-mcp.bridge
     group: level
-    description: "Remove a tag from an actor's Tags array. Params: actorLabel, tag (#219)"
+    description: "Remove a tag from an actor's Tags array. Params: actorLabel OR actorPath, tag (#219)"
     options:
       method: remove_actor_tag
   level.set_actor_tags:
     class_path: ue-mcp.bridge
     group: level
-    description: "Replace an actor's Tags array. Params: actorLabel, tags[] (#219)"
+    description: "Replace an actor's Tags array. Params: actorLabel OR actorPath, tags[] (#219)"
     options:
       method: set_actor_tags
   level.list_actor_tags:
     class_path: ue-mcp.bridge
     group: level
-    description: "List an actor's Tags. Params: actorLabel (#219)"
+    description: "List an actor's Tags. Params: actorLabel OR actorPath (#219/#983)"
     options:
       method: list_actor_tags
   level.attach_actor:
     class_path: ue-mcp.bridge
     group: level
-    description: "Attach actor as child. Params: childLabel, parentLabel, attachRule? (KeepWorld|KeepRelative|SnapToTarget; default KeepWorld), socketName? (#205)"
+    description: "Attach actor as child. Params: childLabel OR childPath, parentLabel OR parentPath, attachRule? (KeepWorld|KeepRelative|SnapToTarget; default KeepWorld), socketName? (#205)"
     options:
       method: attach_actor
   level.detach_actor:
     class_path: ue-mcp.bridge
     group: level
-    description: "Detach actor from parent. Params: childLabel (#205)"
+    description: "Detach actor from parent. Params: childLabel OR childPath (#205/#983)"
     options:
       method: detach_actor
   level.attach_component:
     class_path: ue-mcp.bridge
     group: level
-    description: "Attach an actor root or exact named child SceneComponent to an actor root or exact named parent SceneComponent. A non-root child changes only component hierarchy, not actor parentage. An already-matching attachment is a no-op and does not reapply transform rules. Returns verified resolved component and socket details; native attachment failure is an error. Params: childLabel, parentLabel, childComponentName? (default child root), parentComponentName? (default parent root), socketName?, attachRule? (KeepWorld|KeepRelative|SnapToTarget; default KeepWorld), weldSimulatedBodies? (default false)"
+    description: "Attach an actor root or exact named child SceneComponent to an actor root or exact named parent SceneComponent. A non-root child changes only component hierarchy, not actor parentage. An already-matching attachment is a no-op and does not reapply transform rules. Returns verified resolved component and socket details; native attachment failure is an error. Params: childLabel OR childPath, parentLabel OR parentPath, childComponentName? (default child root), parentComponentName? (default parent root), socketName?, attachRule? (KeepWorld|KeepRelative|SnapToTarget; default KeepWorld), weldSimulatedBodies? (default false)"
     options:
       method: attach_component
   level.detach_component:
     class_path: ue-mcp.bridge
     group: level
-    description: "Detach an actor root or exact named child SceneComponent while preserving world transform. Omitted childComponentName selects the actor root. An already-detached component is a verified no-op. Returns previous parent component/socket details plus alreadyDetached and detachmentChanged. Params: childLabel, childComponentName?"
+    description: "Detach an actor root or exact named child SceneComponent while preserving world transform. Omitted childComponentName selects the actor root. An already-detached component is a verified no-op. Returns previous parent component/socket details plus alreadyDetached and detachmentChanged. Params: childLabel OR childPath, childComponentName?"
     options:
       method: detach_component
   level.set_actor_mobility:
     class_path: ue-mcp.bridge
     group: level
-    description: "Set actor root component Mobility. Params: actorLabel, mobility (static|stationary|movable) (#205)"
+    description: "Set actor root component Mobility. Params: actorLabel OR actorPath, mobility (static|stationary|movable) (#205)"
     options:
       method: set_actor_mobility
   level.get_current_edit_level:
@@ -1600,7 +1730,7 @@ tasks:
   level.batch_translate:
     class_path: ue-mcp.bridge
     group: level
-    description: "Translate a set of actors by an offset. Params: offset (Vec3), actorLabels[] OR tag (#203)"
+    description: "Translate a set of actors by an offset. Params: offset (Vec3), actorLabels[] and/or actorPaths[] OR tag (#203/#983)"
     options:
       method: batch_translate
   level.place_actors_batch:
@@ -1609,16 +1739,106 @@ tasks:
     description: "Bulk-spawn StaticMeshActors with per-instance mesh + transform. Params: actors[]: [{staticMesh, location?, rotation?, scale?, label?}]. Mesh loads cached per path. Returns spawned/failedMesh/failedSpawn counts + labels (#264)"
     options:
       method: place_actors_batch
+  level.batch_set_properties:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Set the same properties on every actor an editor-side SELECTOR picks. batch_translate needs an explicit label list, which for auto-numbered labels is a payload you have to generate first; this takes tag / classFilter / labelPrefix / labelContains / folderPath instead, so '190 SpotLights tagged LichtTest' is one call. Property writes delegate to set_actor_property, so dotted paths, actor-label object references and TArray-of-actor-labels all behave identically to the single-actor action. The whole batch is ONE undo transaction. dryRun resolves every property against every matched actor's class and reports currentValue without writing. The level is left dirty and is NOT saved. Params: properties (object of propertyName -> value, max 32), at least one of actorLabels[]/labelPrefix/labelContains/tag/classFilter/folderPath/folderPathPrefix, matchSubclasses? (default true), dryRun?, force? (bypass EditDefaultsOnly), transactionLabel? (#984)"
+    options:
+      method: batch_set_actor_properties
+  level.bulk_set_component_property:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Set one property on the SAME NAMED COMPONENT across every actor a selector picks (the real case: NavModifierComponent.AreaClass on dozens of placed StaticMeshActors). Delegates to set_component_property per actor, so value semantics match exactly, including JSON null to clear an object reference. An actor that has no component of that name is reported as no_such_component rather than counted as a failure or silently skipped. One undo transaction. The level is left dirty and is NOT saved. Params: componentName, propertyName, value, at least one of actorLabels[]/labelPrefix/labelContains/tag/classFilter/folderPath/folderPathPrefix, matchSubclasses? (default true), dryRun?, transactionLabel? (#941)"
+    options:
+      method: bulk_set_component_property
+  level.remove_components_by_class:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Remove every instance component of a class across the actors a selector picks. remove_component takes one name at a time, which cannot reach 230 orphaned SplineMeshComponents whose names are auto-generated (SplineMeshComponent_0, _1, ...). dryRun DEFAULTS TO TRUE here, unlike delete_actors: this deletes components you cannot enumerate by name, so you see the list first. Only INSTANCE components are removed; a native or SCS component belongs to the class and is reported under skippedNonInstanceComponents rather than destroyed, because destroying it would come back on the next construction rerun. Omitting every actor selector means every loaded actor, which is a legitimate request and the reason for the dryRun default. One undo transaction. Params: componentClass, actorClassFilter? (or classFilter), componentNameContains?, matchComponentSubclasses? (default true), actorLabels?/labelPrefix?/labelContains?/tag?/folderPath?/folderPathPrefix?, dryRun? (default TRUE), save? (default false; the level is otherwise left dirty), transactionLabel? (#907)"
+    options:
+      method: remove_components_by_class
+  level.spawn_actors_batch:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Bulk-spawn actors of ANY class with per-instance properties, from explicit transforms or from positions DERIVED in the editor. place_actors_batch is StaticMeshActor only and spawn_light takes one position. Exactly one source: instances[] ({location?, rotation?, scale?, label?, properties?}); fromComponents ({componentClass?, componentNameContains?, actor selector, offset?, extentFraction?, space?, inheritRotation?}) which places one actor per matched component at boundsOrigin + boxExtent*extentFraction + offset; or alongSpline ({actorLabel, componentName?, spacing (cm), startDistance?, endDistance?, offset?, alignToTangent? (default true)}). fromComponents defaults to space='local' on purpose: a component's WORLD bounds are axis-aligned, so 'the top of the mesh' computed from them is not on a rotated lamp post, while local bounds carry the component's own orientation. That is what makes 'a SpotLight at 72 percent of each of 190 lamp posts' one call instead of 190 bounds reads and 190 spawns. dryRun returns every computed transform without spawning. maxSpawn (default 500) refuses a plan bigger than you meant. The level is left dirty and is NOT saved. Params: actorClass, exactly one of instances[]/fromComponents/alongSpline, properties? (applied to every spawn; per-instance properties win), labelPrefix?, dryRun?, maxSpawn?, transactionLabel? (#987)"
+    options:
+      method: spawn_actors_batch
+  level.set_component_materials:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Set per-slot COMPONENT material overrides on placed actors. This is the other half of material assignment and the half that had no action: material(build_material) and asset(set_property) write the mesh ASSET's slots, while a placed actor usually needs the instance override the details panel edits (OverrideMaterials). set_actor_material does one slot on whichever primitive component it finds first, so dressing a placed skeletal mesh meant a Python loop over get_num_materials(). Exactly one of: materials[] (index = slot; a null or empty entry clears just that slot's override), material (one path applied to EVERY slot), clearOverrides=true (drop every override so the asset's own slots show through). Every slot comes back with its previous and new path AND whether each is a componentOverride or the meshDefault, read back rather than echoed. Every material path is resolved before anything is written, so a bad path fails the call instead of half-applying it. Works on any UMeshComponent (static or skeletal). One undo transaction; the LEVEL is dirtied and NOT saved, the mesh asset is untouched. Params: at least one of actorLabels[]/labelPrefix/labelContains/tag/classFilter/folderPath/folderPathPrefix, componentName? (default the actor's first mesh component), materials? OR material? OR clearOverrides?, dryRun?, transactionLabel? (#946)"
+    options:
+      method: set_component_materials
+  level.spawn_transient_actor:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Spawn an actor for VERIFICATION that cannot be saved into the map. Checking anything that only exists on a live actor needs a subject, and spawning a normal one and forgetting to delete it is how a SceneCapture2D ended up committed to source control. So this actor is transient by construction: RF_Transient (the package serializer skips it), bTemporaryEditorActor (an editor preview actor, kept out of the level's persistent actor list), no external actor package on a World Partition map, and its label is set with bMarkDirty=false because SetActorLabel dirties the map by default. It carries a UEMCP_TransientVerification tag so it can be listed and so destroy refuses anything it did not create, and the response reports dirtiedPackages rather than promising nothing was dirtied. initialize controls how far towards a running state it is taken, because the EDITOR WORLD HAS NOT BEGUN PLAY and a spawned actor therefore gets neither InitializeComponent nor BeginPlay: 'none' spawns only, 'construction' (default) reruns construction scripts, 'beginPlay' also registers components and dispatches the begin-play cycle on this actor, which is what a component doing its setup in BeginPlay (a GAS ability system component's attribute-set scan, for one) needs before it can be read. Each returned component reports registered and initialized so the caller can see what actually happened. Destroy it with destroy_transient_actor. Params: actorClass, location?, rotation?, scale?, label?, initialize? (none|construction|beginPlay), hideFromOutliner?, world? (editor|pie) (#956)"
+    options:
+      method: spawn_transient_actor
+  level.destroy_transient_actor:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Destroy an actor created by spawn_transient_actor. It will ONLY destroy an actor carrying the UEMCP_TransientVerification tag and RF_Transient: a label or path that resolves to a real level actor is refused by name rather than deleted, because this is the cleanup call and a mistyped label at cleanup time is how real work disappears. Use delete_actor or delete_actors for a real level actor. Params: actorPath OR actorLabel OR all=true (every transient verification actor in this world), world? (editor|pie) (#956)"
+    options:
+      method: destroy_transient_actor
+  level.list_transient_actors:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "List the verification actors spawned by spawn_transient_actor that are still in the world, so nothing is left behind by accident. They cannot be saved into the map and do not survive a map reload, but they are live until destroyed. Params: world? (editor|pie) (#956)"
+    options:
+      method: list_transient_actors
+  level.convert_brushes_to_static_mesh:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Convert BSP brushes into StaticMeshActors via UEditorActorSubsystem::ConvertActors, generating the static meshes into destinationPath. Finishing a blockout pass otherwise meant doing it by hand. dryRun DEFAULTS TO TRUE, because conversion DESTROYS the source brushes and there is no way back to a builder brush once the map is saved. Every candidate the selector picked comes back with a verdict, so nothing is silently dropped: the default builder brush, brush shapes, volumes (both derive from ABrush and neither is geometry a player sees) and subtractive brushes (they carve space and have no surface, so converting one yields an empty mesh that looks like success) are all excluded unless you opt in by name. classFilter is EXACT by default here. Reports convertedFrom (recorded before the call, since the actors are gone after it), newActorLabels and newStaticMeshes. Params: actorLabels[] OR folderPath (+ recursiveFolder?, default true), classFilter?, exactClass? (default true), destinationPath? (default /Game/Meshes/Converted), dryRun? (default TRUE), allowSubtractive?, includeVolumes? (#911)"
+    options:
+      method: convert_brushes_to_static_mesh
+  level.rerun_construction:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Re-run the construction script on PLACED actor instances, so an SCS or construction-script change reaches the copies already in the level instead of only new ones. blueprint(run_construction_script) builds a TEMPORARY actor and throws it away, and Python cannot do this at all because Actor.rerun_construction_scripts() is missing from the wrapper. Requires actorLabels or className: this destroys and rebuilds every generated component on each match, so it will not run against a whole level implicitly. Reports reran, failed[], samples[], and how many matches have no user construction script (their generated components are still rebuilt, but a caller expecting a script change to land needs to know). Leaves the level dirty and does NOT save. Params: actorLabels? (string[]) and/or className? (Blueprint or native class), matchSubclasses? (default true), world? (editor|pie) (#944)"
+    options:
+      method: rerun_construction_scripts
+  level.recreate_physics_state:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Rebuild the physics and collision state on the components a selector picks, after the mesh data underneath them changed. Stale collision makes line_trace report a MISS through solid geometry, which is the worst kind of wrong answer because nothing about it looks wrong. dryRun DEFAULTS TO TRUE and is the preflight: it lists which components would be rebuilt and what their collision enabled state and profile are right now. Requires at least one bound (actorLabels, labelPrefix, tag, classFilter, componentClass, componentNameContains) and never refreshes a whole world implicitly. Returns per-component before/after collision state plus a post-refresh collisionSummary histogram. Params: actorLabels?, labelPrefix?, tag?, classFilter?, componentClass?, componentNameContains?, dryRun? (default TRUE), maxComponents? (default 2000), world? (editor|pie) (#915)"
+    options:
+      method: recreate_physics_state
+  level.test_component_overlap:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Do two placed components overlap? Returns the boolean plus both sides' local bounds, world bounds and world transforms, and separation (when apart) or penetration depth (when overlapping). For a UBoxComponent it also returns boxExtentUnscaled, the authored BoxExtent the details panel shows, which no other read exposed. method=OBB (default) runs a separating-axis test on the components' oriented local bounds, so a rotated box is tested as a rotated box; method=AABB compares the axis-aligned world bounds and adds a per-axis gap. Deliberately geometric rather than a physics overlap query: a physics answer depends on collision state, and stale collision state is what makes a trace lie (see recreate_physics_state). Omitting a componentName selects that actor's root component. Params: actorLabelA OR actorPathA, actorLabelB OR actorPathB, componentNameA?, componentNameB?, method? (OBB|AABB), world? (editor|pie) (#914)"
+    options:
+      method: test_component_overlap
   level.line_trace:
     class_path: ue-mcp.bridge
     group: level
-    description: "Line trace in the editor world. Traces simple collision by default, the same as a gameplay trace, so the result matches what the running game hits; pass traceComplex to trace per-triangle instead. channel picks the collision channel, including channels this project renamed in its collision settings. The response echoes traceComplex and channel, and faceIndex is only present on a complex hit. Returns hit + actorLabel/actorClass/componentName/componentClass/location/impactPoint/normal/distance/faceIndex/boneName/physicalMaterial. Params: start (Vec3), end? (Vec3) OR direction? (Vec3) + distance? (default 200000), traceComplex? (default false), channel? (default Visibility), ignoreActors? (array of labels) (#420, #807)"
+    description: "Line trace. world selects which world is traced: 'editor' (default) or 'pie' for the LIVE running PIE world, which is the only way to check geometry that exists once play has begun (#933). Traces simple collision by default, the same as a gameplay trace, so the result matches what the running game hits; pass traceComplex to trace per-triangle instead. channel picks the collision channel, including channels this project renamed in its collision settings. The response echoes traceComplex, channel and the world that actually answered, and faceIndex is only present on a complex hit. Returns hit + actorLabel/actorClass/componentName/componentClass/location/impactPoint/normal/distance/faceIndex/boneName/physicalMaterial. Params: start (Vec3), end? (Vec3) OR direction? (Vec3) + distance? (default 200000), traceComplex? (default false), channel? (default Visibility), ignoreActors? (array of labels), world? (editor|pie), pieInstance? (#420, #807, #933)"
     options:
       method: line_trace
+  level.bulk_line_trace:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Batch of line traces. Each traces[] item uses the same semantics as line_trace (start + end or direction+distance, simple collision by default, optional traceComplex/channel/ignoreActors). The top-level world applies to the WHOLE batch, so one call cannot straddle the editor and PIE worlds and present the results as comparable. Caps at 256. Returns results[] in traces[] order, one per item, each with the same fields as line_trace (or success=false + error for that item). Params: traces[], world? (editor|pie), pieInstance? (#933)"
+    options:
+      method: bulk_line_trace
+  level.snap_instances_to_surface:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Project ISMC/HISMC instances down onto surface geometry, preserving each instance's rotation and scale. Dry-run by default: preflights every requested instance and reports what would move before anything does, then applies successful projections in one transaction with rollback. Filter the surfaces traced against with surfaceActorClass/surfaceActorLabels so instances do not snap onto each other. onMiss controls whether a miss aborts the batch or is skipped. Does not save packages. Params: actorLabel OR actorPath, componentName?, instanceIndices? (omit for all), maxInstances?, traceStartOffset?, traceDistance?, surfaceOffset?, channel?, traceComplex?, surfaceActorClass?, surfaceActorLabels?, onMiss? ('abort'|'skip'), dryRun? (default true) (#905)"
+    options:
+      method: snap_instances_to_surface
+  level.summarize_static_mesh_usage:
+    class_path: ue-mcp.bridge
+    group: level
+    description: "Read-only inventory of which static meshes are actually placed in the loaded world, aggregated in one game-thread scan instead of walking every actor and component. This is the action for placement counts, and query_components is not a substitute for it: grouping there only increments, so one HISM holding 5,000 instances counts as 1, while this SUMS real ISM/HISM instances and counts distinct actors per mesh, and it sorts every aggregate before truncating so the top N is the true top N with an exact totalUniqueMeshes. Distinguishes actors, components and placements. Results are bounded; full-scan totals are preserved alongside the truncated rows. Loads nothing and saves nothing. Params: world? (editor|pie), maxResults?, includeOccurrences?, maxOccurrences? (#903)"
+    options:
+      method: summarize_static_mesh_usage
   level.snap_actor_to_floor:
     class_path: ue-mcp.bridge
     group: level
-    description: "Snap an actor's bounds-bottom to the first downward line-trace hit. Equivalent of the End-key shortcut, works on arbitrary geometry (not just Landscape). Params: actorLabel, floorOffset? (added to impact Z, default 0), maxDistance? (default 100000) (#419)"
+    description: "Snap an actor's bounds-bottom to the first downward line-trace hit. Equivalent of the End-key shortcut, works on arbitrary geometry (not just Landscape). The actor lookup and the downward trace both run in the world named by world, so a PIE snap measures against PIE geometry (#933). Params: actorLabel OR actorPath, floorOffset? (added to impact Z, default 0), maxDistance? (default 100000), world? (editor|pie), pieInstance? (#419, #933)"
     options:
       method: snap_actor_to_floor
 
@@ -1626,19 +1846,19 @@ tasks:
   material.read:
     class_path: ue-mcp.bridge
     group: material
-    description: "Read material structure. Params: assetPath"
+    description: "Read material structure. Accepts a Material or a MaterialInstance; an instance answers with its lineage, shading setup and resolved parameters instead of a graph (#952). Params: assetPath"
     options:
       method: read_material
   material.list_parameters:
     class_path: ue-mcp.bridge
     group: material
-    description: "List overridable parameters. Params: assetPath"
+    description: "List parameters. On a MaterialInstance each entry carries the resolved value, the parent's defaultValue and whether this instance overrides it (#952). Params: assetPath"
     options:
       method: list_material_parameters
   material.set_parameter:
     class_path: ue-mcp.bridge
     group: material
-    description: "Set parameter on MaterialInstance. Params: assetPath, parameterName, parameterType, value"
+    description: "Set parameter on MaterialInstance. Params: assetPath, parameterName, parameterType? (scalar|vector|texture, auto-detected when omitted), value, association?. A vector/colour value may be passed as value OR color, shaped {r,g,b,a}, [r,g,b,a] or '(R=..,G=..,B=..,A=..)'. Saves the instance and returns readBack + saved so a set/verify round trip needs no second call (#952)"
     options:
       method: set_material_parameter
   material.read_instance:
@@ -1680,7 +1900,7 @@ tasks:
   material.set_expression_value:
     class_path: ue-mcp.bridge
     group: material
-    description: "Set a value on an expression node. Common typed knobs: value (constant/param default), texturePath (TextureSample/TextureSampleParameter2D default texture). For any OTHER UPROPERTY on the expression (e.g. SamplerType, SamplerSource, Group), pass propertyName + value to hit the generic reflection setter. Params: materialPath, expressionIndex, value?, texturePath?, propertyName? (#663)"
+    description: "Set a value on an expression node. Common typed knobs: value (constant/param default), texturePath (TextureSample/TextureSampleParameter2D default texture). Constant2Vector/3Vector/4Vector and VectorParameter all take value (or color) as {r,g,b,a}, {x,y,z,w}, [r,g,b,a] or '(R=..,G=..)'; Constant2Vector reads the first two components (#979). Saves the material and reports saved. For any OTHER UPROPERTY on the expression (e.g. SamplerType, SamplerSource, Group), pass propertyName + value to hit the generic reflection setter. Params: materialPath, expressionIndex, value?, color?, texturePath?, propertyName? (#663)"
     options:
       method: set_expression_value
   material.set_custom_expression:
@@ -1731,6 +1951,12 @@ tasks:
     description: "List expression nodes inside a MaterialFunction. Params: functionPath (#463)"
     options:
       method: list_expressions_in_function
+  material.build_material:
+    class_path: ue-mcp.bridge
+    group: material
+    description: "Build a PBR material from a texture set in one call: creates a TextureSample per entry, picks each sampler type FROM THE TEXTURE (a virtual/UDIM texture silently loses its material connection on recompile unless it samples through the Virtual* variant), wires them, recompiles, saves, and reports which connections survived the recompile. Params: name + packagePath? (create) OR materialPath (build into an existing material); textures = {baseColor, normal, roughness, metallic, specular, emissive, opacity, opacityMask, ambientOcclusion, ...} plus packed keys orm (R=AO,G=Roughness,B=Metallic) and rma; samplerTypes? = per-key override (e.g. {normal: 'VirtualNormal'}); clearExisting?; assignToMesh? (StaticMesh or SkeletalMesh asset) with meshSlots? (names or indices, default all) (#946)"
+    options:
+      method: build_material
   material.create_simple:
     class_path: ue-mcp.bridge
     group: material
@@ -1958,7 +2184,7 @@ tasks:
   animation.add_notify:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Add notify. For PlayMontageNotify the notifyName is also written onto the spawned notify object so OnPlayMontageNotifyBegin broadcasts it (not 'None'), and montage branching-point markers refresh (#528). notifyProperties writes EditAnywhere fields onto the spawned notify object and therefore requires a notifyClass that resolves. Params: assetPath, notifyName, triggerTime, notifyClass?, notifyProperties?"
+    description: "Add notify. For PlayMontageNotify the notifyName is also written onto the spawned notify object so OnPlayMontageNotifyBegin broadcasts it (not 'None'), and montage branching-point markers refresh (#528/#880). On a montage the PlayMontageNotify classes are added as BRANCHING POINT notifies, which is the only tick type UAnimNotify_PlayMontageNotify::BranchingPointNotify runs at, so OnPlayMontageNotifyBegin broadcasts without a montage reload; pass branchingPoint to force it either way, and read branchingPointMarkerCount in the response to see what the montage cached. notifyProperties writes EditAnywhere fields onto the spawned notify object and therefore requires a notifyClass that resolves. Params: assetPath, notifyName, triggerTime, notifyClass?, notifyProperties?"
     options:
       method: add_anim_notify
   animation.remove_notify:
@@ -2144,9 +2370,15 @@ tasks:
   animation.read_ik_rig:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Read IK Rig chains, solvers, skeleton. Params: assetPath"
+    description: "Read an IK Rig's preview mesh, skeleton roots/bones, ancestry-validated chains and goal assignments, concrete goals, exclusions, and structured solver/FBIK effector state. Params: assetPath"
     options:
       method: read_ik_rig
+  animation.configure_ik_rig:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "UE 5.8 only. Author an existing IK Rig through UIKRigController with strict bone, ancestry, goal, and setting validation, native readback, one transaction, and checked save; older engines return unsupported_engine_version. autoSetup='retarget' installs the native retarget definition; 'full_body' installs the retarget definition then Full Body IK before requested desired-state upserts. Params: rigPath, autoSetup? ('retarget'|'full_body'), retargetRoot?, rootMotionBone?, chains?: [{name,startBone,endBone,goal?}], fullBodyIK?: {solverIndex?,rootBone,enabled?,goals:[{name,bone,positionAlpha?,rotationAlpha?,chainDepth?,strengthAlpha?,pullChainAlpha?,pinRotation?}]}, exclusions?: [{bone,excluded}]."
+    options:
+      method: configure_ik_rig
   animation.list_control_rig_variables:
     class_path: ue-mcp.bridge
     group: animation
@@ -2165,6 +2397,36 @@ tasks:
     description: "Read a Control Rig's per-element hierarchy metadata: each element's name, type (Bone|Control|Null|Curve...), index, and parent. Params: assetPath (#619)"
     options:
       method: read_control_rig_hierarchy
+  animation.begin_control_rig_edit:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "UE 5.8 only. Create a Sequencer Control Rig editing session over a source AnimSequence; native returns unsupported_engine_version on older engines. Baseline first: before this call, reuse or create a Control Rig for the target character, bind/import the exact target skeleton, add the intended controls, author Forward Solve, add Backward/Inverse Solve, verify it with read_control_rig_hierarchy/read_control_rig_graph, and pass an unchanged source round-trip. For a new baseline, the bundled Epic 5.8 controlrig actions include epic_create, epic_import_bones_from_asset, epic_add_control, and epic_add_backward_solve_graph; epic_create alone is not a usable rig. There is no silent fallback to raw bone-key authoring. rigMode='fk' uses UFKControlRig only when generated FK controls are sufficient; rigMode='asset' requires the verified controlRigPath and rejects rigs without inverse execution. bindingTag is the stable natural key for replay. onConflict is skip|error (default error); existing sessions are never modified. layered defaults false. startFrame is inclusive and endFrame is exclusive. Params: sequencePath, skeletalMeshPath, sourceAnimationPath, rigMode ('fk'|'asset'), controlRigPath?, layered?, startFrame?, endFrame?, displayRate?, bindingTag?, onConflict?. Returns the resolved bindingTag/binding GUID, rig, frame range, controls and created/existed status."
+    options:
+      method: begin_control_rig_edit
+  animation.read_control_rig_edit:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "UE 5.8 only. Read transform, bool, float/scale-float, and integer/enum controls from a Control Rig editing session without changing editor state; native returns unsupported_engine_version on older engines and has no silent fallback. Params: sequencePath, bindingTag, controlNames?, frames?, space? ('local'|'global'). Scalar samples return value instead of transform. Control metadata includes native controlType, animatable, and enum path/options where applicable. Returns session identity, layered mode, range/rate, filtered control metadata, and requested frame samples."
+    options:
+      method: read_control_rig_edit
+  animation.apply_control_rig_edits:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "UE 5.8 only. Apply typed Control Rig edits in one transaction; native returns unsupported_engine_version on older engines. There is no silent fallback to raw bone tracks. set_keys writes strictly ordered full per-frame transforms from normalized quaternions and preserves shortest-arc quaternion continuity. A set operation writes one full absolute transform at frame or frames. An offset operation applies translation/rotation/scale deltas across an inclusive frame range with optional edge blends. contact_lock densely constrains a translatable driver control, or an optional driven bone/socket reference, to a fixed component-space target with smooth edge blends and optional pole/control stabilization. Driver and stabilizer keys are read back transactionally. A drivenReference contact returns verification='bake_and_analyze_required'; bake it and analyze every constrained frame before accepting the bone/socket result. set_bool, set_float, and set_int key matching scalar controls; enum controls use set_int with one of the integer values reported in enumOptions. Params: sequencePath, bindingTag, operations[] where set_keys={op:'set_keys',control,keys:[{frame,transform:{translation,rotationQuaternion,scale}}],space?}, set={op:'set',control,frame|frames,transform:{translation,rotationDegrees,scale},space?}, offset={op:'offset',control,startFrame,endFrame,translationCm?,rotationDegrees?,scaleMultiplier?,space?,blendInFrames?,blendOutFrames?}, contact_lock={op:'contact_lock',control,drivenReference?,startFrame,endFrame,target:{translation,rotationQuaternion?},blendInFrames?,blendOutFrames?,stabilizeControls?,positionToleranceCm?,rotationToleranceDegrees?}, set_bool={op:'set_bool',control,frame|frames,value}, set_float={op:'set_float',control,frame|frames,value}, or set_int={op:'set_int',control,frame|frames,value}. Sequencer's current interpolation mode is retained. Returns per-operation counts, affected controls/frames, and contactQa summaries; a failed key/readback batch is undone."
+    options:
+      method: apply_control_rig_edits
+  animation.bake_control_rig_edit:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "UE 5.8 only. Bake the evaluated Control Rig session to a new AnimSequence asset; native returns unsupported_engine_version on older engines and has no raw-track fallback. The source LevelSequence remains unchanged. outputAssetPath is the output natural key; onConflict is skip|error (default error), never overwrite. Key reduction and Sequencer links are not supported yet, so reduceKeys/createLink must be false or omitted. Params: sequencePath, bindingTag, outputAssetPath, frameRate?, reduceKeys?, tolerance?, createLink?, onConflict?. Returns output asset metadata, frame/rate counts, status, and delete-created-asset rollback."
+    options:
+      method: bake_control_rig_edit
+  animation.analyze_animation:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "Cross-version, data-driven AnimSequence inspection using the native animation APIs available in the compiled engine. Samples an AnimSequence and reports deterministic numeric motion diagnostics without Python or viewport inference. Params: assetPath (required AnimSequence), skeletalMeshPath?, boneNames?, frames?, sampleRate?, loop?, outputDirectory? (must resolve under Project/Saved/Codex/AnimationQA and must not already contain artifacts). Returns source/rate/range metadata, sampled local/component transforms, root-motion and continuity metrics, and any written analysis artifacts."
+    options:
+      method: analyze_animation
   animation.set_root_motion:
     class_path: ue-mcp.bridge
     group: animation
@@ -2204,9 +2466,15 @@ tasks:
   animation.read_ik_retargeter:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Read IKRetargeter: source/target rigs and chain mappings. Params: assetPath (#246)"
+    description: "Read an IK Retargeter's source/target rigs and preview meshes, flattened and per-op chain mappings, typed op stack, and all named/current pose offsets when the compiled engine exposes them. Params: assetPath (#246)"
     options:
       method: read_ik_retargeter
+  animation.configure_ik_retargeter:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "UE 5.8 only. Configure an existing IK Retargeter through UIKRetargeterController with the correct default-op and per-op rig assignment order, auto/manual chain mappings, named pose authoring, processor validation, native readback, transaction rollback, and checked save; older engines return unsupported_engine_version. Whole-pose auto-align resets that pose first: create a new pose or pass pose.reset=true to acknowledge replacement, then manual offsets are applied. Params: retargeterPath, sourceRig?, targetRig?, sourcePreviewMesh?, targetPreviewMesh?, ensureDefaultOps? (default true), autoMapMode? ('exact'|'fuzzy'|'clear'), forceRemap? (default false), chainMappings?: [{targetChain,sourceChain?:string|null}], pose?: {side,name,create?,reset?,autoAlign?,bones?,rotationOffsets?:[{bone,rotationQuaternion}],rootOffsetZ?,snapBoneToGround?}."
+    options:
+      method: configure_ik_retargeter
   animation.set_ik_rig_mesh:
     class_path: ue-mcp.bridge
     group: animation
@@ -2234,7 +2502,7 @@ tasks:
   animation.batch_retarget_animations:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Bake a set of source AnimSequences onto the target skeleton through an IK Retargeter (RunBatchRetarget). Params: retargeterPath, sourceMesh, targetMesh, animPaths[], outputPath? (default: alongside source), prefix?, suffix? (default _Retargeted), overwrite? (#701)"
+    description: "Bake validated source AnimSequences onto the target skeleton through an IK Retargeter (RunBatchRetarget), save every output, and roll back newly created outputs if the batch is incomplete or unsavable. Overwrite is rejected. Returns mapping completeness and every unmapped target chain so partial retargets are explicit; pass requireCompleteMapping=true only when the target should have no intentional extra chains. Params: retargeterPath, sourceMesh, targetMesh, animPaths[], outputPath? (default: alongside source), prefix?, suffix? (default _Retargeted), overwrite? (must be false), requireCompleteMapping? (default false) (#701)"
     options:
       method: batch_retarget_animations
   animation.set_anim_blueprint_skeleton:
@@ -2378,25 +2646,43 @@ tasks:
   animation.get_bone_transform:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Read a bone or socket transform on a live actor's SkeletalMeshComponent. Wraps GetBoneTransform / GetSocketTransform. Params: actorLabel, boneName (or socket name), componentName? (default: CharacterMesh0 / Mesh / first SK component), world? (auto|pie|game|editor, default auto), space? (world|component|local, default world). Returns location, rotation, scale (#420)"
+    description: "Read a bone or socket transform on a live actor's SkeletalMeshComponent. Wraps GetBoneTransform / GetSocketTransform. Params: actorLabel OR actorPath, boneName (or socket name), componentName? (default: CharacterMesh0 / Mesh / first SK component), world? (auto|pie|game|editor, default auto), space? (world|component|local, default world). Returns location, rotation, scale (#420)"
     options:
       method: get_bone_transform
   animation.list_bones:
     class_path: ue-mcp.bridge
     group: animation
-    description: "List bones in a live actor's SkeletalMeshComponent ref skeleton (name, index, parent). Params: actorLabel, componentName?, world? (auto|pie|game|editor, default auto) (#420)"
+    description: "List bones in a live actor's SkeletalMeshComponent ref skeleton (name, index, parent). Params: actorLabel OR actorPath, componentName?, world? (auto|pie|game|editor, default auto) (#420)"
     options:
       method: list_bones
   animation.rebind_leader_pose:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Re-bind every secondary SkeletalMeshComponent on an actor to a body component (default CharacterMesh0 / Mesh). One-call fix for the 'character explodes after rotating the actor' failure mode. Params: actorLabel, bodyComponent? (#419)"
+    description: "Re-bind every secondary SkeletalMeshComponent on an actor to a body component (default CharacterMesh0 / Mesh). One-call fix for the 'character explodes after rotating the actor' failure mode. Params: actorLabel OR actorPath, bodyComponent? (#419)"
     options:
       method: rebind_leader_pose
+  animation.sample_pose:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "Evaluate an AnimSequence (or a BlendSpace at a blend position) and return COMPOSED bone transforms, which read_bone_track (raw local space) and get_bone_transforms (reference pose only) cannot give. Params: assetPath, boneNames? (omit for every bone), frames? or times? (omit for every sampled key), space? (component default | local | world), skeletalMeshPath? (evaluate with that mesh's proportions), incorporateRootMotion? (default true), blendPosition? {x,y,z} (BlendSpace only). Returns samples:[{frame, time, bones:[{name, location, rotation, scale}]}]. A montage is refused with a pointer to its segments, because a montage has no pose of its own (#923/#926/#922)"
+    options:
+      method: sample_pose
+  animation.get_live_bone_transforms:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "Read the EVALUATED pose off a live SkeletalMeshComponent in the editor world or PIE, for many bones at once. This is the read that tells 'standing' from 'prone' when the reference pose cannot (#922). Also returns componentTransform and an evaluation block (animationMode, animInstanceClass, componentSpaceTransformCount) so a clean component transform sitting on a dead anim instance is visible in one call. Params: actorLabel OR actorPath, componentName? (default CharacterMesh0 / Mesh), boneNames? (omit for every bone, max 1000), space? (world default | component | local), world? (auto|pie|game|editor) (#922/#926)"
+    options:
+      method: get_live_bone_transforms
+  animation.measure_natural_speed:
+    class_path: ue-mcp.bridge
+    group: animation
+    description: "Measure a locomotion clip's planted-foot speed, in cm/s, by evaluating the pose and tracking the lowest foot's horizontal travel while it is in contact. Every retarget changes natural speed by the target skeleton's leg-length ratio, so a BlendSpace built on retargeted clips has to re-measure per clip per character (#923). Params: assetPath, footBones[] (e.g. ['foot_l','foot_r']), contactThreshold? (contact height in cm; omit to derive it from the clip), skeletalMeshPath?, frames?/times?, blendPosition? (BlendSpace only). Returns naturalSpeed, plantedDistance, plantedTime, contactThreshold, per-foot breakdown"
+    options:
+      method: measure_natural_speed
   animation.preview_animation:
     class_path: ue-mcp.bridge
     group: animation
-    description: "Toggle bUpdateAnimationInEditor + VisibilityBasedAnimTickOption=AlwaysTickPoseAndRefreshBones on every SkeletalMeshComponent of an actor. Bypasses the 'cannot be edited on templates' guard for level instances. Params: actorLabel, enabled (#419/#420)"
+    description: "Toggle bUpdateAnimationInEditor + VisibilityBasedAnimTickOption=AlwaysTickPoseAndRefreshBones on every SkeletalMeshComponent of an actor. Bypasses the 'cannot be edited on templates' guard for level instances. Params: actorLabel OR actorPath, enabled (#419/#420)"
     options:
       method: preview_animation
 
@@ -2416,19 +2702,19 @@ tasks:
   landscape.sample:
     class_path: ue-mcp.bridge
     group: landscape
-    description: "Sample height/layers. Params: x, y"
+    description: "Read the surface height AND every paint-layer weight at one world XY. Height and weights come from the landscape's own height and weight data (the merged result of every edit layer), the same data landscape(paint_layer) writes, so they are exact and do not depend on built collision. Layers come back with weight (0..1), weight255 (the weightmap byte the editor shows), the LayerInfo path and its physical material, plus dominantLayer and totalWeight - which is what makes reading a paint weight one call instead of rendering the weightmap to a render target and reading that back. Also returns quad, quadExtent and inBounds, so a zero weight OFF the landscape reads as out of bounds rather than as a measured zero, and traceHeight/normal from a confirmation collision trace (absent when collision is unbuilt or the proxy is streamed out). Position accepts x + y, point {x, y}, or worldX + worldY. Params: x + y (or point, or worldX + worldY), actorLabel? OR actorPath? (which landscape, when the map has several), layerName? (one layer only), includeLayers? (default true) (#939)"
     options:
       method: sample_landscape
   landscape.sculpt:
     class_path: ue-mcp.bridge
     group: landscape
-    description: "Raise, lower or flatten terrain with a circular brush. mode=raise|lower|flatten; amount is world centimetres at full strength (raise/lower), and flatten pulls toward the height under the brush centre. falloff (0..1) is the smoothstepped soft edge. Runs in one transaction and leaves the level dirty and unsaved. Writes into an edit layer (editLayer by name, else editLayerIndex, default 0) - required on UE 5.8, where a write without one is regenerated away by the layer system. Params: center ({x,y} world space), radius (default 500), mode? (default raise), amount? (default 100), falloff? (default 0.5), actorLabel?, editLayer?, editLayerIndex?, maxVertices? (#742)"
+    description: "Raise, lower or flatten terrain with a circular brush. mode=raise|lower|flatten; amount is world centimetres at full strength (raise/lower), and flatten pulls toward the height under the brush centre. falloff (0..1) is the smoothstepped soft edge. Runs in one transaction and leaves the level dirty and unsaved. Writes into an edit layer (editLayer by name, else editLayerIndex, default 0) - required on UE 5.8, where a write without one is regenerated away by the layer system. Params: center ({x,y} world space), radius (default 500), mode? (default raise), amount? (default 100), falloff? (default 0.5), actorLabel? OR actorPath?, editLayer?, editLayerIndex?, maxVertices? (#742)"
     options:
       method: sculpt_landscape
   landscape.paint_layer:
     class_path: ue-mcp.bridge
     group: landscape
-    description: "Paint a weight layer with a circular brush. The layer must already have a LayerInfo on this landscape (see add_layer_info) - an unregistered name errors and lists the layers that ARE registered instead of silently painting nothing. Weights are written as given - the engine no longer renormalises other layers for you, so set them explicitly if they must sum to 1. Writes into an edit layer (editLayer by name, else editLayerIndex, default 0). Params: layerName, center ({x,y} world space), radius (default 500), strength? (0..1, default 1), falloff? (default 0.5), actorLabel?, editLayer?, editLayerIndex?, maxVertices? (#742)"
+    description: "Paint a weight layer with a circular brush. The layer must already have a LayerInfo on this landscape (see add_layer_info) - an unregistered name errors and lists the layers that ARE registered instead of silently painting nothing. Weights are written as given - the engine no longer renormalises other layers for you, so set them explicitly if they must sum to 1. Writes into an edit layer (editLayer by name, else editLayerIndex, default 0). Params: layerName, center ({x,y} world space), radius (default 500), strength? (0..1, default 1), falloff? (default 0.5), actorLabel? OR actorPath?, editLayer?, editLayerIndex?, maxVertices? (#742)"
     options:
       method: paint_landscape_layer
   landscape.list_splines:
@@ -2485,6 +2771,12 @@ tasks:
     description: "Resolve which loaded LandscapeStreamingProxy covers a world X/Y. Returns found/loaded + label, or loaded:false when the covering proxy is streamed out (so a 0-weight readback there is ambiguous, not real). Params: worldX, worldY (#733)"
     options:
       method: find_landscape_proxy_at
+  landscape.refresh_physical_material_collision:
+    class_path: ue-mcp.bridge
+    group: landscape
+    description: "UE 5.8+: safely refresh physical-material collision data in memory on loaded World Partition LandscapeStreamingProxy actors after a LayerInfo PhysMaterial change. Requires complete registered collision coverage and no pending landscape edit-layer work. Preserves and verifies every raw, complex-live, and simple-live height sample, builds material data before one collision recreation, and fails the whole matched batch on any unsafe result. Filters combine: actorLabels[], guids[], and bounds {min,max}; omitting them targets every loaded proxy up to maxActors (default 256, hard max 1024). Unloaded proxies are untouched; pin them first with level(load_actor_descs). Refuses PIE/SIE and non-World-Partition maps. Persistence is deliberately unsupported because Landscape PreSave can mutate edit-layer collision data; this action never saves packages. Returns loaded/matched/refreshed/failed counts and exact affected package paths."
+    options:
+      method: refresh_landscape_physical_material_collision
 
   # ── pcg ──
   pcg.list_graphs:
@@ -2514,7 +2806,7 @@ tasks:
   pcg.get_component_details:
     class_path: ue-mcp.bridge
     group: pcg
-    description: "Inspect PCG component. Params: actorLabel"
+    description: "Inspect PCG component. Params: actorLabel OR actorPath (#983)"
     options:
       method: get_pcg_component_details
   pcg.create_graph:
@@ -2562,25 +2854,25 @@ tasks:
   pcg.execute:
     class_path: ue-mcp.bridge
     group: pcg
-    description: "Regenerate PCG. Params: actorLabel"
+    description: "Regenerate PCG. Params: actorLabel OR actorPath (#983)"
     options:
       method: execute_pcg_graph
   pcg.force_regenerate:
     class_path: ue-mcp.bridge
     group: pcg
-    description: "Force a stuck PCG component to regenerate (clears graph ref, re-sets, cleanup+generate). Params: actorLabel (#146)"
+    description: "Force a stuck PCG component to regenerate (clears graph ref, re-sets, cleanup+generate). Params: actorLabel OR actorPath (#146/#983)"
     options:
       method: force_regenerate_pcg
   pcg.cleanup:
     class_path: ue-mcp.bridge
     group: pcg
-    description: "Cleanup a PCG component (remove spawned content). Params: actorLabel, removeComponents? (default true) (#146)"
+    description: "Cleanup a PCG component (remove spawned content). Params: actorLabel OR actorPath, removeComponents? (default true) (#146)"
     options:
       method: cleanup_pcg
   pcg.toggle_graph:
     class_path: ue-mcp.bridge
     group: pcg
-    description: "Toggle a PCG component's graph assignment to force reinit (no generate). Params: actorLabel, graphPath? (#146)"
+    description: "Toggle a PCG component's graph assignment to force reinit (no generate). Params: actorLabel OR actorPath, graphPath? (#146)"
     options:
       method: toggle_pcg_graph
   pcg.add_volume:
@@ -2630,9 +2922,15 @@ tasks:
   foliage.set_settings:
     class_path: ue-mcp.bridge
     group: foliage
-    description: "Modify type settings. Params: foliageTypeName, settings"
+    description: "Modify ONE foliage type's settings. For the same edit across every type matching a condition, use batch_set_settings_where instead of calling this in a loop. Params: foliageTypeName OR foliageTypePath, settings"
     options:
       method: set_foliage_type_settings
+  foliage.batch_set_settings_where:
+    class_path: ue-mcp.bridge
+    group: foliage
+    description: "Write settings to every FoliageType whose CURRENT property values match a predicate, with the predicate evaluated in the editor. set_settings is one asset per call and cannot express a condition at all, so 'include_in_hlod=false on every type whose cull distance is set' meant reading 201 assets out, computing the 36 matches client-side, and writing them back one at a time. Candidates come from foliageTypePaths[] (explicit), directory (scan FoliageType assets) or fromLevel=true (types placed in the open level). Predicate fields read the asset's own properties by dotted path, so CullDistance.Max works and a bare name is treated the same as props.<name>; operators are the same vocabulary as level(query_components). dryRun DEFAULTS TO TRUE because this writes to many assets at once, and the preview still resolves every setting path so an unwritable property is reported before the commit. Written values are read back rather than echoed, so a value the property coerced is visible. Params: settings (object of propertyName -> value, dotted paths supported), where ([{field, op, value}], required), whereMode? (all|any), one of foliageTypePaths[]/directory (+ recursive?)/fromLevel, propertyNames? (extra properties to report without filtering on them), dryRun? (default TRUE), save? (default true), maxTypes? (#988)"
+    options:
+      method: batch_set_foliage_settings_where
 
   # ── niagara ──
   niagara.list:
@@ -2668,13 +2966,13 @@ tasks:
   niagara.reactivate:
     class_path: ue-mcp.bridge
     group: niagara
-    description: "Reset + reactivate the NiagaraComponent on a placed actor (replay a burst before capturing). Params: actorLabel (#537)"
+    description: "Reset + reactivate the NiagaraComponent on a placed actor (replay a burst before capturing). Params: actorLabel OR actorPath (#537/#983)"
     options:
       method: reactivate_niagara
   niagara.set_parameter:
     class_path: ue-mcp.bridge
     group: niagara
-    description: "Set parameter. Params: actorLabel, parameterName, value, parameterType?"
+    description: "Set parameter. Params: actorLabel OR actorPath, parameterName, value, parameterType?"
     options:
       method: set_niagara_parameter
   niagara.create:
@@ -3218,11 +3516,11 @@ tasks:
   editor.start_editor:
     class_path: editor.start_editor
     group: editor
-    description: "Launch Unreal Editor and BLOCK until it is fully ready (not merely until the socket answers), rendering a startup progress bar in the terminal. Returns the phase timeline it waited through. Do NOT poll get_engine_state or get_status afterwards: this call already waited, and a ready editor is the only way it returns success. Params: timeout? (seconds, default 300)"
+    description: "Launch Unreal Editor and BLOCK until it is fully ready (not merely until the socket answers), rendering a startup progress bar in the terminal. Returns the phase timeline it waited through. Do NOT poll get_engine_state or get_status afterwards: this call already waited, and a ready editor is the only way it returns success. dialogPolicy answers startup prompts before they can wedge the game thread, which is how the post-crash \"Restore Packages\" modal used to stall a launch (#968). Params: timeout? (seconds, default 300), dialogPolicy? (\"pattern=response;pattern=response\", responses as set_dialog_policy takes them)"
   editor.get_engine_state:
     class_path: editor.get_engine_state
     group: editor
-    description: "What the engine is REALLY doing, read from outside the game thread: startup phase from the editor's own log, process table (PID, command line, responding), the plugin's status snapshot (slow-task name and percent, active modal dialog, game-thread stall), and native dialog windows. Call this ONCE when something is already wrong (handlers timing out, an editor that will not come up). Never call it in a wait loop: start_editor blocks until ready on its own, and polling this during startup burns tokens re-reading state that is already tracked. Params: probeWindows? (default true; scans native windows, costs ~2s)"
+    description: "What the engine is REALLY doing, read from outside the game thread: startup phase from the editor's own log, every process holding this project's .uproject open (PID, command line, responding), the plugin's status snapshot (slow-task name and percent, active modal dialog, game-thread stall), and native dialog windows. `running` follows the strongest evidence: an editor that answered over the bridge is running whatever the process table saw, and a probe that could not run is reported as processProbeFailed rather than as an absent editor (#965). Call this ONCE when something is already wrong (handlers timing out, an editor that will not come up). Never call it in a wait loop: start_editor blocks until ready on its own, and polling this during startup burns tokens re-reading state that is already tracked. Params: probeWindows? (default true; scans native windows, costs ~2s)"
   editor.stop_editor:
     class_path: editor.stop_editor
     group: editor
@@ -3244,7 +3542,7 @@ tasks:
   editor.execute_python:
     class_path: editor.execute_python
     group: editor
-    description: "GATED LAST RESORT. execute_python is unreachable until a semantic tool search over your taskSummary has been run AND every candidate it returns is EXPLICITLY ruled out with a stated reason. Flow: (1) call with taskSummary (+code) - it returns the candidate actions; (2) re-call with the same taskSummary/code PLUS ruledOut=[{action, reason}] giving a specific reason each candidate does not fit. Python runs only once every candidate is ruled out. Params: code, taskSummary (required), ruledOut?, resultVariable? (name of a top-level variable to return as `result`, separate from print()/log; #732) (#704)"
+    description: "GATED LAST RESORT. execute_python is unreachable until a semantic tool search over your taskSummary has been run AND every candidate it returns is EXPLICITLY ruled out with a stated reason. Flow: (1) call with taskSummary (+code) - it returns the candidate actions AND the exact ruledOut array to send back; (2) re-call with the same taskSummary/code PLUS that ruledOut=[{action, reason}], each reason at least 12 characters saying why that candidate does not fit. The action field accepts the bare name, tool(action) or tool.action, and rulings are remembered for the session so rewording the taskSummary never asks you to justify the same action twice. Python runs only once every candidate is ruled out. Params: code, taskSummary (required), ruledOut?, resultVariable? (name of a top-level variable to return as `result`, separate from print()/log; #732) (#704, #938, #960)"
   editor.run_python_file:
     class_path: ue-mcp.bridge
     group: editor
@@ -3306,7 +3604,7 @@ tasks:
   editor.get_runtime_value:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Read PIE actor property. Params: actorLabel, propertyName (supports dotted paths: component.field or component.struct.field for nested reads on component subobjects, #344/#381)"
+    description: "Read PIE actor property. Params: actorLabel OR actorPath, propertyName (supports dotted paths: component.field or component.struct.field for nested reads on component subobjects, #344/#381)"
     options:
       method: get_runtime_value
   editor.get_pie_pawn:
@@ -3324,7 +3622,7 @@ tasks:
   editor.invoke_object_function:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Call a UFUNCTION on any UObject, not just a placed actor. Target it with objectPath, or target=gameinstance|gamemode|gamestate|playercontroller|playerpawn|subsystem (subsystem also needs subsystemClass; playercontroller/playerpawn accept playerIndex). The GameInstance, GameMode and subsystems have no actor label, so invoke_function could never reach them. Returns output and return params under returnValues; an unknown function name lists the available ones. Params: functionName, objectPath? | target?, subsystemClass?, playerIndex?, args?, world? (editor|pie|auto), pieInstance? (#739)"
+    description: "Call a UFUNCTION on any UObject, not just a placed actor. Target it with objectPath, or target=gameinstance|gamemode|gamestate|playercontroller|playerpawn|subsystem (subsystem also needs subsystemClass; playercontroller/playerpawn accept playerIndex). The GameInstance, GameMode and subsystems have no actor label, so invoke_function could never reach them. Returns output and return params under returnValues, with a TArray/TSet/TMap return as real JSON and everything else as export text (#885); an unknown function name lists the available ones. A scripted call runs under the editor script-execution guard, which forces every actor callspace to Local, so a UFUNCTION(Server) executes locally instead of being sent; the result warns when that happened, and deferToNextTick=true queues the call for the next engine tick where it routes normally, at the cost of returning before it runs (#973). Params: functionName, objectPath? | target?, subsystemClass?, playerIndex?, args?, world? (editor|pie|auto), pieInstance?, deferToNextTick? (#739)"
     options:
       method: invoke_object_function
   editor.invoke_object_functions:
@@ -3336,7 +3634,7 @@ tasks:
   editor.read_bone_transforms:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Read live skeletal bone and socket transforms off an actor, once. This is a point-in-time read, NOT a time series - for per-frame capture over a window use the pie category's observe actions. Pass bones (bone OR socket names) or omit for every bone up to limit. space=world (default) or component; component space is independent of where the actor is standing. Pass relativeTo (bone OR socket name) to express every sample in that live reference frame; relativeTo supersedes space and is calculated in component space. Also reports the AnimInstance class/path. Params: actorLabel, componentName?, bones?, relativeTo?, space?, limit?, world? (editor|pie|auto), pieInstance? (#756/#757/#761/#764)"
+    description: "Read live skeletal bone and socket transforms off an actor, once. This is a point-in-time read, NOT a time series - for per-frame capture over a window use the pie category's observe actions. Pass bones (bone OR socket names) or omit for every bone up to limit. space=world (default) or component; component space is independent of where the actor is standing. Pass relativeTo (bone OR socket name) to express every sample in that live reference frame; relativeTo supersedes space and is calculated in component space. Also reports the AnimInstance class/path. Params: actorLabel OR actorPath, componentName?, bones?, relativeTo?, space?, limit?, world? (editor|pie|auto), pieInstance? (#756/#757/#761/#764)"
     options:
       method: read_bone_transforms
   editor.get_object_properties:
@@ -3348,7 +3646,7 @@ tasks:
   editor.set_movement_mode:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Set a live PIE character's movement mode and/or velocity on its CharacterMovementComponent. Modes are named (none|walking|navwalking|falling|swimming|flying|custom) rather than raw enum numbers, because a wrong number reads as success and then behaves as None. Reports previousMode/previousVelocity and reads the mode back afterwards, since SetMovementMode can refuse a mode the character cannot enter (flying with bCanFly off, swimming outside a volume). Params: actorLabel, mode?, customMode? (only with mode='custom'), velocity? {x,y,z}, world? (default pie), pieInstance? (#757)"
+    description: "Set a live PIE character's movement mode and/or velocity on its CharacterMovementComponent. Modes are named (none|walking|navwalking|falling|swimming|flying|custom) rather than raw enum numbers, because a wrong number reads as success and then behaves as None. Reports previousMode/previousVelocity and reads the mode back afterwards, since SetMovementMode can refuse a mode the character cannot enter (flying with bCanFly off, swimming outside a volume). Params: actorLabel OR actorPath, mode?, customMode? (only with mode='custom'), velocity? {x,y,z}, world? (default pie), pieInstance? (#757)"
     options:
       method: set_movement_mode
   editor.set_object_property:
@@ -3366,19 +3664,19 @@ tasks:
   editor.teleport_runtime_actor:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Move a live PIE actor and have it STAY moved. A plain SetActorLocation on a Character is undone by CharacterMovement on the next tick, so this stops the movement component, teleports, and stops it again. Reports actualLocation read back from the actor rather than what was requested. Params: actorLabel, location?, rotation?, stopMovement? (default true), sweep? (default false), world? (default pie), pieInstance? (#770/#777)"
+    description: "Move a live PIE actor and have it STAY moved. A plain SetActorLocation on a Character is undone by CharacterMovement on the next tick, so this stops the movement component, teleports, and stops it again. Reports actualLocation read back from the actor rather than what was requested. Params: actorLabel OR actorPath, location?, rotation?, stopMovement? (default true), sweep? (default false), world? (default pie), pieInstance? (#770/#777)"
     options:
       method: teleport_runtime_actor
   editor.invoke_static_function:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Call a static UFUNCTION on a UBlueprintFunctionLibrary (no actor instance). invoke_function needs an actor/component target; this targets the library class CDO instead, so it reaches static *_BlueprintOnly libraries (Voxel sculpt/query/stamp), GeometryScript, Kismet math, any function library. Params: className (short name or /Script/Module.Class path), functionName, args? (name -> JSON value, same marshalling as invoke_function), actorArgs? (name -> actor label for UObject* params that are actors, e.g. the sculpt actor), worldContextParam? (name of a UObject* param to fill with the editor/PIE world; auto-detected for params named WorldContextObject), world? (editor|pie). Returns return/out params under returnValues. Discover libraries + functions with list_function_libraries."
+    description: "Call a static UFUNCTION on a UBlueprintFunctionLibrary (no actor instance). invoke_function needs an actor/component target; this targets the library class CDO instead, so it reaches static *_BlueprintOnly libraries (Voxel sculpt/query/stamp), GeometryScript, Kismet math, any function library. Params: className (short name or /Script/Module.Class path), functionName, args? (name -> JSON value, same marshalling as invoke_function), actorArgs? (name -> actor label for UObject* params that are actors, e.g. the sculpt actor), worldContextParam? (name of a UObject* param to fill with the selected world; auto-detected from the function's own WorldContext metadata and for params named WorldContextObject), world? (editor|pie|game|auto, default editor), pieInstance? (which PIE world; see list_pie_instances). world/pieInstance pick WHICH world fills the context param, so a static that looks a GameInstance subsystem up off its context can be exercised against a live PIE session (#971); the result names the world it ran against and the parameter that carried it. Returns return/out params under returnValues, with a TArray/TSet/TMap return as real JSON and everything else as export text (#885). Discover libraries + functions with list_function_libraries."
     options:
       method: invoke_static_function
   editor.invoke_function:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Call a BlueprintCallable / Exec UFUNCTION on a target actor or one of its components. world=editor (the default) runs the function on the actor placed in the level, no PIE session needed, and reports which instance ran it as resolvedActorLabel/resolvedActorPath. actorLabel is matched against placed actors by editor label first, then internal object name, then full object path; a miss is an error naming what was searched (#806). Params: actorLabel, functionName, component? (component subobject name; redirects target from the actor to that component, #382), args? (object; struct values accept a JSON object such as {X,Y,Z} as well as an export-text string), actorArgs? (object mapping UObject* parameter name to actor label, resolved against live actors in the active world; #383), world? (editor|pie). Returns out/return params (#228/#229)"
+    description: "Call a BlueprintCallable / Exec UFUNCTION on a target actor or one of its components. world=editor (the default) runs the function on the actor placed in the level, no PIE session needed, and reports which instance ran it as resolvedActorLabel/resolvedActorPath. actorLabel is matched against placed actors by editor label first, then internal object name, then full object path; a miss is an error naming what was searched (#806). Returns out/return params under returnValues; a TArray/TSet/TMap return comes back as real JSON and everything else as export text (#885). A scripted call runs under the editor script-execution guard, which forces every actor callspace to Local, so a UFUNCTION(Server) executes locally instead of being sent; the result warns when that happened, and deferToNextTick=true queues the call for the next engine tick where it routes normally, at the cost of returning before it runs (#973). Params: actorLabel OR actorPath, functionName, component? (component subobject name; redirects target from the actor to that component, #382), args? (object; struct values accept a JSON object such as {X,Y,Z} as well as an export-text string), actorArgs? (object mapping UObject* parameter name to actor label, resolved against live actors in the active world; #383), world? (editor|pie), deferToNextTick? (#228/#229)"
     options:
       method: invoke_function
   editor.list_function_libraries:
@@ -3438,13 +3736,13 @@ tasks:
   editor.capture_screenshot:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Screenshot. target=pie synchronously captures the selected PIE client game viewport with UMG/Slate UI; target=editor captures the level viewport; target=window synchronously captures a whole Slate window via FSlateApplication::TakeScreenshot - pixel-true for ALL Slate/UMG UI, returns after the PNG is written, and works while the window is unfocused or off-screen. Multi-instance PIE automatically prefers a world with a game viewport; pass pieInstance or worldPath to select explicitly, which for target=window is what picks the PIE client window instead of the active editor window. Every mode captures at the source viewport/window size - use capture_scene_png for a chosen output size. Params: filename?, target? (auto|pie|editor|window), pieInstance?, worldPath?. Returns the resolved PIE instance/world and image dimensions (#226/#724)."
+    description: "Screenshot. target=pie synchronously captures the selected PIE client game viewport with UMG/Slate UI; target=editor captures the level viewport; target=window synchronously captures a whole Slate window via FSlateApplication::TakeScreenshot - pixel-true for ALL Slate/UMG UI, returns after the PNG is written, and works while the window is unfocused or off-screen. Multi-instance PIE automatically prefers a world with a game viewport; pass pieInstance or worldPath to select explicitly, which for target=window is what picks the PIE client window instead of the active editor window. Every mode captures at the source viewport/window size - use capture_scene_png for a chosen output size. Params: filename? (outputPath is accepted for it too, so the two capture actions take the same name; #966), target? (auto|pie|editor|window), pieInstance?, worldPath?. Returns the resolved PIE instance/world and image dimensions (#226/#724)."
     options:
       method: capture_screenshot
   editor.capture_scene_png:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Headless PNG screenshot via SceneCapture2D (works unfocused, guaranteed RGBA8 LDR). focusActorLabel auto-frames the camera on an actor's bounds; world:pie captures the running game world (#599). Params: outputPath, location?, rotation?, focusActorLabel?, focusDirection?, focusMargin?, world? (editor|pie), width? (default 1280), height? (default 720), fov? (default 90) (#148/#599)"
+    description: "Headless PNG screenshot via SceneCapture2D (works unfocused, guaranteed RGBA8 LDR). focusActorLabel auto-frames the camera on an actor's bounds; world:pie captures the running game world (#599). The capture actor is transient and destroyed before the call returns, and any left by an earlier build is swept, so a capture leaves the level exactly as it found it (#966). Params: outputPath (filename is accepted for it too), location?, rotation?, focusActorLabel? OR focusActorPath, focusDirection?, focusMargin?, world? (editor|pie), width? (default 1280), height? (default 720), fov? (default 90) (#148/#599)"
     options:
       method: capture_scene_png
   editor.set_realtime:
@@ -3468,7 +3766,7 @@ tasks:
   editor.get_runtime_values:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Bulk runtime read across the active world. For each actor/component matching classFilter, resolves every path against the (actor|component) root and returns rows of {actorLabel, actorClass, componentName?, componentClass?, values, errors?}. Paths support property hops, sub-object hops, and zero-arg BlueprintCallable getter calls at any segment (e.g. 'PowerConnector.GetRequired' reaches a UFUNCTION on a UObject sub-object). classFilter matches actor class OR component class - omit to match everything. World defaults to PIE if running, else editor. Params: classFilter?, paths[], world? (editor|pie) (#414)"
+    description: "Bulk runtime read across the active world. For each actor/component matching classFilter, resolves every path against the (actor|component) root and returns rows of {actorLabel, actorClass, componentName?, componentClass?, values, errors?}. Paths support property hops, sub-object hops, and BlueprintCallable getter calls at any segment (e.g. 'PowerConnector.GetRequired' reaches a UFUNCTION on a UObject sub-object). A getter that takes arguments is written with them inline, 'GetMirroredTallyWeight(overclock)' or 'GetBalance(gold, 2)', which is what makes a keyed accessor readable across every matched instance in one call; the literals are coerced by the same rules invoke_object_function's args use, so FName/FString/int/float/bool/enum all read (#969). classFilter matches actor class OR component class - omit to match everything. A path whose result is a TArray/TSet/TMap comes back as real JSON rather than one string (#885). World defaults to PIE if running, else editor. Params: classFilter?, paths[], world? (editor|pie) (#414)"
     options:
       method: get_runtime_values
   editor.set_viewport:
@@ -3480,7 +3778,7 @@ tasks:
   editor.focus_on_actor:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Focus on actor. Params: actorLabel"
+    description: "Focus on actor. Params: actorLabel OR actorPath (#983)"
     options:
       method: focus_viewport_on_actor
   editor.create_sequence:
@@ -3492,25 +3790,25 @@ tasks:
   editor.get_sequence_info:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Read sequence: bindings (possessable/spawnable) with their Sequencer tags (#556), tracks, and optional section detail. Params: assetPath, includeSectionDetails? (attach sockets, first transform key values per track)"
+    description: "Read sequence: bindings (possessable/spawnable) with their Sequencer tags (#556), tracks, and optional section detail. UNITS: playbackRange is reported in TICKS (tick resolution, commonly 24000/s), while MovieSceneScripting*Channel.add_key defaults its time_unit to DISPLAY_RATE. Keys authored with a tick number under the default unit land roughly 800x past the range and the track evaluates to its first key, which presents as transforms that do not work on a structurally perfect sequence (#881). Params: assetPath, includeSectionDetails? (attach sockets, first transform key values per track)"
     options:
       method: get_sequence_info
   editor.add_sequence_track:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Add an empty track. Params: assetPath, trackType, actorLabel?"
+    description: "Add an empty track. Params: assetPath, trackType, actorLabel? OR actorPath? (#983)"
     options:
       method: add_sequence_track
   editor.add_sequence_section:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Add a section to a track (creating the track if needed), set its start/end in seconds, and for a CameraCut track bind it to a camera. Returns the section index + channel names to key. Params: sequencePath, trackType (Transform|Float|Fade|CameraCut|Audio|Event|SkeletalAnimation), actorLabel? (binding scope), startSeconds?, endSeconds?, cameraActorLabel? (#548)"
+    description: "Add a section to a track (creating the track if needed), set its start/end in seconds, and for a CameraCut track bind it to a camera. Returns the section index + channel names to key. Params: sequencePath, trackType (Transform|Float|Fade|CameraCut|Audio|Event|SkeletalAnimation), actorLabel? OR actorPath? (binding scope), startSeconds?, endSeconds?, cameraActorLabel? OR cameraActorPath? (#548/#983)"
     options:
       method: add_sequence_section
   editor.set_sequence_keyframes:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Add keyframes to a section channel. Transform channels: Location.X/Y/Z, Rotation.X/Y/Z (or friendly x/y/z, yaw/pitch/roll); Fade/Float: the float channel. Params: sequencePath, trackType, actorLabel?, sectionIndex? (default 0), channel, keyframes ([{seconds, value}]), interpolation? (cubic|linear) (#548)"
+    description: "Add keyframes to a section channel. Transform channels: Location.X/Y/Z, Rotation.X/Y/Z (or friendly x/y/z, yaw/pitch/roll); Fade/Float: the float channel. Params: sequencePath, trackType, actorLabel? OR actorPath?, sectionIndex? (default 0), channel, keyframes ([{seconds, value}]), interpolation? (cubic|linear) (#548)"
     options:
       method: set_sequence_keyframes
   editor.set_sequence_playback_range:
@@ -3525,6 +3823,12 @@ tasks:
     description: "Play/stop/pause a Level Sequence in Sequencer. Pass sequencePath (or assetPath) to target a specific sequence - it is opened first, because the underlying Sequencer commands act on whatever is currently open. Omit it and the call applies to the open sequence and says so. Params: sequencePath? (or assetPath), sequenceAction? (play|pause|stop, default play)"
     options:
       method: play_sequence
+  editor.scrub_sequence:
+    class_path: ue-mcp.bridge
+    group: editor
+    description: "Park the Sequencer playhead on an exact time and evaluate there, then return. This is what makes scrub-then-capture_scene_png deterministic: play_sequence only offers play/pause/stop and realtime playback races the capture. Pauses first, scrubs, and forces the evaluation before answering, because the playhead move alone does not write possessed-actor transforms. Pass exactly one of seconds or frame. frame is read in timeUnit: display (default, the frame numbers Sequencer shows) or tick (the units get_sequence_info's playbackRange reports). Returns the evaluated time in all three units plus displayRate/tickResolution/playbackRange, and warns when the time is outside the playback range. Params: sequencePath? (or assetPath), seconds? | frame?, timeUnit? (#881)"
+    options:
+      method: scrub_sequence
   editor.build_all:
     class_path: ue-mcp.bridge
     group: editor
@@ -3672,7 +3976,7 @@ tasks:
   editor.run_automation_tests:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Run registered Automation tests matching a filter and return per-test pass/fail plus error lines. Runs them synchronously through the test framework rather than the console queue, and suspends the editor's unfocused-CPU throttle for the duration - otherwise an unfocused editor drops to a few FPS and the framework's interactive-frame-rate gate never opens, leaving tests queued forever (#765). Params: filter?, maxTests? (default 50) (#693)"
+    description: "Run registered Automation tests matching a filter and return per-test pass/fail plus error lines. Runs them synchronously through the test framework rather than the console queue, and suspends the editor's unfocused-CPU throttle for the duration - otherwise an unfocused editor drops to a few FPS and the framework's interactive-frame-rate gate never opens, leaving tests queued forever (#765). A test whose latent commands are still queued when latentTimeoutSeconds runs out is reported as abandoned, with the reason: latent work needs engine frames, and this runs on the game thread, so a test that starts PIE (CQTest multi-client network tests, for instance) belongs in the editor's Automation window or -ExecCmds=\"Automation RunTests <name>\" at launch. Such a test used to terminate the editor outright (#993). Params: filter?, maxTests? (default 50), latentTimeoutSeconds? (default 5, max 120) (#693)"
     options:
       method: run_automation_tests
   editor.list_dirty_packages:
@@ -3681,12 +3985,16 @@ tasks:
     description: "Enumerate currently-dirty content + map packages, read from the editor's own dirty-package lists (the same ones Save All uses). Includes a never-saved /Temp world, because an unsaved new map is exactly the unsaved work a caller needs to see before closing or reloading (#340)"
     options:
       method: list_dirty_packages
-  editor.request_editor_shutdown:
+  editor.get_world_state:
     class_path: ue-mcp.bridge
     group: editor
-    description: "Ask the editor to close itself from inside the engine, after it has checked that closing is safe. Refuses by default when any content or map package is dirty (including an unsaved /Temp world) and reports which ones, so nothing is lost to a silent discard. Ends an active PIE/SIE session first and closes only once play has actually stopped. The response is returned before the process exits. Use editor(stop_editor) for the full stop-and-confirm flow; this action is the in-engine half of it. Params: requireClean? (default true), endPIE? (default true)"
+    description: "One atomic read of which world is open and what is unsaved. Returns editorWorldName, editorWorldPackage, persistentLevelPackage, worldPackageDirty, a sorted dirtyPackages list with counts, and the editor/play/simulate mode. level(get_current) plus editor(list_dirty_packages) is two calls, so the editor can change between them and neither result proves which world the other described; this answers both in one game-thread dispatch. Read-only, and fails closed rather than reporting an empty world as a clean one (#920/#921)"
     options:
-      method: request_editor_shutdown
+      method: get_world_state
+  editor.request_editor_shutdown:
+    class_path: editor.request_editor_shutdown
+    group: editor
+    description: "Ask the editor to close itself from inside the engine, after it has checked that closing is safe. Refuses by default when any content or map package is dirty (including an unsaved /Temp world) and reports which ones, so nothing is lost to a silent discard. Ends an active PIE/SIE session first and closes only once play has actually stopped. The response is returned before the process exits. Aimed at the same editor stop_editor aims at, through the same ownership check, so the two can never disagree about which editor belongs to the loaded project (#967). Use editor(stop_editor) for the full stop-and-confirm flow; this action is the in-engine half of it. Params: requireClean? (default true), endPIE? (default true)"
 
   # ── reflection ──
   reflection.reflect_class:
@@ -3766,37 +4074,37 @@ tasks:
   gameplay.set_collision_profile:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Set collision preset. Params: actorLabel, profileName"
+    description: "Set collision preset. Params: actorLabel OR actorPath, profileName"
     options:
       method: set_collision_profile
   gameplay.set_simulate_physics:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Toggle physics. Params: actorLabel, simulate"
+    description: "Toggle physics. Params: actorLabel OR actorPath, simulate"
     options:
       method: set_simulate_physics
   gameplay.add_impulse:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Apply an impulse (or force with mode='force') to a (PIE) actor's simulating physics body so its motion can be observed over time - loop level(read_actor_motion) to sample the response. Params: actorLabel, impulse {x,y,z} (or force/vector), mode? (impulse|force), componentName?, boneName?, location? (apply at world point), velChange?/accelChange?, world? (editor|pie|auto) (#676)"
+    description: "Apply an impulse (or force with mode='force') to a (PIE) actor's simulating physics body so its motion can be observed over time - loop level(read_actor_motion) to sample the response. Params: actorLabel OR actorPath, impulse {x,y,z} (or force/vector), mode? (impulse|force), componentName?, boneName?, location? (apply at world point), velChange?/accelChange?, world? (editor|pie|auto) (#676)"
     options:
       method: add_impulse
   gameplay.set_collision_enabled:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Set collision mode. Params: actorLabel, collisionEnabled"
+    description: "Set collision mode. Params: actorLabel OR actorPath, collisionEnabled"
     options:
       method: set_collision_enabled
   gameplay.set_collision:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Unified collision authoring for a placed actor (actorLabel) or a Blueprint component template (assetPath+componentName). Apply any of: collisionProfile, collisionEnabled (NoCollision|QueryOnly|PhysicsOnly|QueryAndPhysics), objectType (channel name), responseToAllChannels (Block|Overlap|Ignore), responses ({channel: Block|Overlap|Ignore}). Profile is applied first, then overrides. componentName optional for actors (defaults to all primitive components), required for Blueprint templates (#545)"
+    description: "Unified collision authoring for a placed actor (actorLabel or actorPath) or a Blueprint component template (assetPath+componentName). Apply any of: collisionProfile, collisionEnabled (NoCollision|QueryOnly|PhysicsOnly|QueryAndPhysics), objectType (channel name), responseToAllChannels (Block|Overlap|Ignore), responses ({channel: Block|Overlap|Ignore}). Profile is applied first, then overrides. componentName optional for actors (defaults to all primitive components), required for Blueprint templates (#545)"
     options:
       method: set_collision
   gameplay.set_physics_properties:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Set mass/damping/gravity. Params: actorLabel, mass?, linearDamping?, angularDamping?, enableGravity?"
+    description: "Set mass/damping/gravity. Params: actorLabel OR actorPath, mass?, linearDamping?, angularDamping?, enableGravity?"
     options:
       method: set_physics_properties
   gameplay.rebuild_navigation:
@@ -3808,7 +4116,7 @@ tasks:
   gameplay.find_nav_path:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Synchronous nav-path query between two world points. Returns valid/partial/length plus the polyline. The standard 'why doesn't my AI move?' diagnostic. Params: start (Vec3), end (Vec3), pathfindingContext? (actorLabel - uses its agent + filter) (#424)"
+    description: "Synchronous nav-path query between two world points. Returns valid/partial/length plus the polyline. The standard 'why doesn't my AI move?' diagnostic. Params: start (Vec3), end (Vec3), pathfindingContext? (actor label) OR pathfindingContextPath? (object path) - uses its agent + filter (#424/#983)"
     options:
       method: find_nav_path
   gameplay.list_nav_invokers:
@@ -3910,15 +4218,63 @@ tasks:
   gameplay.get_behavior_tree_info:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Inspect behavior tree (top-level + blackboard). Params: assetPath"
+    description: "Inspect a BehaviorTree asset: root node, blackboard asset, every blackboard key (name, type, instanceSynced, description) plus the keys inherited from the blackboard's parent chain, and the node/decorator counts. Params: assetPath (#887)"
     options:
       method: get_behavior_tree_info
   gameplay.read_behavior_tree_graph:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Walk BT tree: composites, tasks, decorators, services with blackboard keys. Params: assetPath"
+    description: "Walk a BT asset: composites, tasks, decorators, services, each with a stable nodePath ('Root.Children[0].Decorators[1]') you pass to read_bt_node_properties / set_bt_node_property. Every decorator reports its full runtime config - blackboardKey {selectedKeyName, selectedKeyType}, basicOperation (Set|NotSet), arithmeticOperation, textOperation, flowAbortMode, notifyObserver, inverseCondition, int/float/stringValue and staticDescription - which is what answers 'the tree is running but the wrong branch fires'. Set includeProperties for each node's own UPROPERTY values, propertyNames to narrow those, includeInherited to keep UBTNode's own fields. Params: assetPath, includeProperties?, includeInherited?, propertyNames? (#888)"
     options:
       method: read_behavior_tree_graph
+  gameplay.read_bt_node_properties:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Filtered read of BT node UPROPERTY values: pick nodes by nodeClass, nodeName, nodePath or kind and get back only their own properties, instead of an unfiltered whole-asset dump. Narrow further with propertyNames. A UE 5.8 FValueOrBBKey_* field (AcceptableRadius, WaitTime, FilterClass) is unpacked into {defaultValue, key, isBound, baseClass, text}. Params: assetPath, nodeClass?, nodeName?, nodePath?, kind?, propertyNames?, includeInherited? (#919)"
+    options:
+      method: read_bt_node_properties
+  gameplay.list_bt_tasks:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Inventory BTTask nodes with their name, class, parent path and FilterClass. Pass assetPath for one tree, or directory (+recursive) to sweep many. On UE 5.8 BTTask_MoveTo.FilterClass is a FValueOrBBKey_Class, so it is reported as {defaultValue, defaultValueName, key, isBound, baseClass, text} - the shape get_editor_property reads as empty. filterClassOnly keeps only tasks that declare one. Params: assetPath?, directory?, recursive?, taskClass?, filterClassOnly?, limit? (#940)"
+    options:
+      method: list_bt_tasks
+  gameplay.set_bt_node_property:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Scoped write onto one owned BT node subobject. Pick the node with nodePath (from read_behavior_tree_graph), nodeName, nodeClass or kind - the write refuses to run unless exactly one node matches - then set property + value, or a properties map of dotted/indexed paths ('Lines[1].Text', 'BlackboardKey.SelectedKeyName'). A scalar aimed at a UE 5.8 FValueOrBBKey_* field lands on its DefaultValue; an object writes the struct's own fields, so {Key: 'MyKey'} binds it to a blackboard key. Every target is read back and reported with its previous value. Params: assetPath, nodePath?, nodeName?, nodeClass?, kind?, property?, value?, properties? (#919)"
+    options:
+      method: set_bt_node_property
+  gameplay.list_bt_graph_nodes:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "List the EDITOR-GRAPH nodes of a BehaviorTree, which is the layer add_bt_node / move_bt_node / remove_bt_node address. Each entry carries its guid (the primary key: stable across recompiles and unaffected by inserting a sibling), category (root|composite|task|decorator|service), class, parentGuid, indexInParent, childGuids in execution order, decoratorGuids, serviceGuids, graph position and any node error. runtimePath cross-references the same node's read address ('Root.Children[0]') so one listing serves both surfaces. A tree that has never been opened in the BT editor reports hasGraph false and no nodes; add_bt_node seeds the graph. Params: assetPath (#889)"
+    options:
+      method: list_bt_graph_nodes
+  gameplay.add_bt_node:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Add a composite, task, decorator or service to a BehaviorTree's editor graph and recompile it into the runnable tree. This is the authoring half that reflection cannot reach: UBehaviorTreeGraph::UpdateAsset, which turns the graph into the UBTCompositeNode tree, is editor-only and unexposed, so nodes spawned by script alone never become a valid tree. parent takes a guid from list_bt_graph_nodes, a runtime address from read_behavior_tree_graph, a unique node name, or 'root' (the default). Decorators and services attach as subnodes of their parent rather than as children. index places the node among its siblings, which is the execution order a Selector or Sequence is defined by; on a SimpleParallel it picks the output (0 = main task, 1 = background). properties writes UPROPERTY values (a bare number aimed at a UE 5.8 FValueOrBBKey_* field lands on its DefaultValue), and blackboardKeys points FBlackboardKeySelector fields at named blackboard keys, resolved against the tree's blackboard. Params: assetPath, nodeClass, nodeCategory?, parent?, index?, nodeName?, properties?, blackboardKeys? (#889)"
+    options:
+      method: add_bt_node
+  gameplay.move_bt_node:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Reconnect a BehaviorTree node to a different parent, reorder it among its siblings, or both, then recompile. Pass parent to reparent and index to reorder; at least one is required. Child order is what a Selector or Sequence executes by, so this is behaviour rather than layout. Decorators and services move between their owners' subnode lists. Moving a node under itself or its own descendant is refused, and the whole subtree travels with it. Returns the parent's siblingOrder after the compile. Params: assetPath, node, parent?, index? (#947)"
+    options:
+      method: move_bt_node
+  gameplay.remove_bt_node:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Delete a node from a BehaviorTree's editor graph and recompile. A composite takes its whole branch with it, so the compiled tree never reports nodes that no longer run. Decorators and services detach from their owner. The root node cannot be removed. Returns every removed node's guid and class. Params: assetPath, node (#889)"
+    options:
+      method: remove_bt_node
+  gameplay.set_bt_task_property:
+    class_path: ue-mcp.bridge
+    group: gameplay
+    description: "Write a property on one BT task node, FilterClass included. Same selector and dotted-path rules as set_bt_node_property. Retarget a MoveTo filter with property='FilterClass' and a class path, a short class name or a Blueprint asset path; pass null to clear it, or {Key: 'MyKey'} to bind it to a blackboard key. The new value is read back and returned. Params: assetPath, nodePath?, nodeName?, nodeClass?, property?, value?, properties? (#940)"
+    options:
+      method: set_bt_task_property
   gameplay.create_blackboard:
     class_path: ue-mcp.bridge
     group: gameplay
@@ -3994,7 +4350,7 @@ tasks:
   gameplay.get_state_tree_runtime:
     class_path: ue-mcp.bridge
     group: gameplay
-    description: "Read a running StateTreeComponent's active state names in PIE (the 'brain' state). Params: actorLabel, world? (default pie), componentName? (#654)"
+    description: "Read a running StateTreeComponent's active state names in PIE (the 'brain' state). Params: actorLabel OR actorPath, world? (default pie), componentName? (#654)"
     options:
       method: get_state_tree_runtime
   gameplay.create_state_tree:
@@ -4170,33 +4526,45 @@ tasks:
   gas.apply_effect:
     class_path: ue-mcp.bridge
     group: gas
-    description: "Apply a GameplayEffect to a live actor's ASC (agnostic stat/damage stimulus - uses the game's own effect). Params: actorLabel, effectClass (content path or class name), level?, setByCaller? ({tag-or-name: magnitude}), world? (auto|pie|editor, default auto)"
+    description: "Apply a GameplayEffect to a live actor's ASC (agnostic stat/damage stimulus - uses the game's own effect). Params: actorLabel OR actorPath, effectClass (content path or class name), level?, setByCaller? ({tag-or-name: magnitude}), world? (auto|pie|editor, default auto)"
     options:
       method: apply_effect
   gas.set_attribute:
     class_path: ue-mcp.bridge
     group: gas
-    description: "Set a gameplay attribute's base value on a live actor's ASC (recalculates CurrentValue through the aggregator). Params: actorLabel, attribute (Health | SetName.Health), value, world?"
+    description: "Set a gameplay attribute's base value on a live actor's ASC (recalculates CurrentValue through the aggregator). Params: actorLabel OR actorPath, attribute (Health | SetName.Health), value, world?"
     options:
       method: set_attribute
   gas.get_attribute:
     class_path: ue-mcp.bridge
     group: gas
-    description: "Read gameplay attribute base + current values on a live actor's ASC. Omit attribute to list all. Params: actorLabel, attribute?, world?"
+    description: "Read gameplay attribute base + current values on a live actor's ASC. Omit attribute to list all. Params: actorLabel OR actorPath, attribute?, world?"
     options:
       method: get_attribute
   gas.init_asc:
     class_path: ue-mcp.bridge
     group: gas
-    description: "Initialize a live actor's ASC (InitAbilityActorInfo) and optionally instantiate an AttributeSet so attributes are live - the runtime setup step for testing a bridge-authored GAS actor. Params: actorLabel, attributeSet? (content path or class name), world?"
+    description: "Initialize a live actor's ASC (InitAbilityActorInfo) and optionally instantiate an AttributeSet so attributes are live - the runtime setup step for testing a bridge-authored GAS actor. Params: actorLabel OR actorPath, attributeSet? (content path or class name), world?"
     options:
       method: init_asc
   gas.get_asc_state:
     class_path: ue-mcp.bridge
     group: gas
-    description: "Introspect a live actor's ASC: granted ability specs (class, level, inputID, active, dynamicTags) + owned gameplay tags. Params: actorLabel, world? (auto|pie|editor) (#587)"
+    description: "Introspect a live actor's ASC: granted ability specs (class, level, inputID, active, dynamicTags) + owned gameplay tags. Params: actorLabel OR actorPath, world? (auto|pie|editor) (#587)"
     options:
       method: get_asc_state
+  gas.get_live_attribute_value:
+    class_path: ue-mcp.bridge
+    group: gas
+    description: "Read the live value of one FGameplayAttributeData on the attribute set instance actually REGISTERED on an actor's AbilitySystemComponent - equivalent to ASC->GetSet<T>(), not the actor's own subobject pointer. Works in the editor world, where no set is registered yet, by first registering the actor's own sets the way BeginPlay would (set registerOwnerSets=false for a strict read). Returns currentValue and baseValue off the instance plus the aggregator's view of both, and the instance's object path so you can prove which object was read. Params: actorLabel (label or internal name) OR actorPath, attributeSet (content path or class name), attribute (property name, or Set.Property), registerOwnerSets?, world? (#956)"
+    options:
+      method: get_live_attribute_value
+  gas.set_live_attribute_value:
+    class_path: ue-mcp.bridge
+    group: gas
+    description: "Write the live value of one FGameplayAttributeData on the REGISTERED attribute set instance on an actor's AbilitySystemComponent. valueType=\"current\" (default) writes the attribute data in place, which is what staging a mid-combat state needs; valueType=\"base\" writes through the ASC so the aggregator recomputes the current value, which is what a durable change needs. The set's PreAttributeChange may clamp, so the result reports what was actually stored alongside the previous values and the instance's object path. Params: actorLabel (label or internal name) OR actorPath, attributeSet, attribute, value, valueType?, registerOwnerSets?, world? (#956)"
+    options:
+      method: set_live_attribute_value
 
   # ── networking ──
   networking.set_replicates:
@@ -4296,7 +4664,7 @@ tasks:
   feedback.submit:
     class_path: feedback.submit
     group: feedback
-    description: "Submit feedback about a tool gap (missing action, wrong behavior, crash, or a case you had to work around). Provide a specific title and a summary; pythonWorkaround and idealTool are optional enrichment, not prerequisites. Checks the plugin registry first and files against the owning plugin's repo when one matches, then blocks on an MCP elicitation prompt that asks the USER (not the agent) to approve or decline the exact payload - and to override the tracker - before anything is posted to GitHub."
+    description: "Submit feedback about a tool gap (missing action, wrong behavior, crash, or a case you had to work around). Provide a specific title and a summary; pythonWorkaround and idealTool are optional enrichment, not prerequisites. Checks the plugin registry first and files against the owning plugin's repo when one matches, then blocks on an MCP elicitation prompt that asks the USER (not the agent) to approve or decline the exact payload - and to override the tracker - before anything is posted to GitHub. If the client cannot show that form (it never advertised elicitation, it throws, or it auto-answers in milliseconds without rendering anything), nothing is lost: the report is written to disk and the result carries a prefilled GitHub issue URL for the user to click."
   feedback.route:
     class_path: feedback.route
     group: feedback