@runpod/mcp-server 4.6.0 → 4.8.0

package/dist/http.mjs

@@ -1,3 +1,10 @@
+import {
+  DIRECTORY_PATH,
+  directoryIncludes,
+  directoryOmittedInputs,
+  profileForPath
+} from "./chunk-6IETGSON.mjs";
+
 // src/http.ts
 import { StreamableHTTPServerTransport } from "@modelcontextprotocol/sdk/server/streamableHttp.js";
 import {
@@ -480,7 +487,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -543,9 +550,36 @@ var generatedTools = [
         },
         ClusterMounts: {
           description: "Storage mounts for a cluster create: `persistent` or `network`, not both. Clusters cannot mount global volumes, so `global` is rejected.",
+          "x-go-type": "BaseMounts",
           allOf: [
             {
               $ref: "#/$defs/BaseMounts"
+            },
+            {
+              type: "object",
+              properties: {
+                persistent: {
+                  type: "object",
+                  properties: {
+                    path: {
+                      type: "string",
+                      minLength: 1
+                    }
+                  }
+                },
+                network: {
+                  type: "array",
+                  items: {
+                    type: "object",
+                    properties: {
+                      path: {
+                        type: "string",
+                        minLength: 1
+                      }
+                    }
+                  }
+                }
+              }
             }
           ],
           unevaluatedProperties: false
@@ -683,7 +717,7 @@ var generatedTools = [
               type: "string",
               description: "ECR resource ARN",
               examples: [
-                "arn:aws:ecr:us-east-2:418399314813:repository/runpod/deployment"
+                "arn:aws:ecr:us-east-2:123456789012:repository/runpod/deployment"
               ]
             },
             name: {
@@ -767,7 +801,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -1018,7 +1052,7 @@ var generatedTools = [
           type: "object",
           required: ["volumeId"],
           additionalProperties: false,
-          description: "A volume mounted on every worker, always at `/runpod-volume`. Custom paths are not supported yet.",
+          description: "A volume mounted on every worker, always at `/runpod-volume`.",
           properties: {
             volumeId: {
               type: "string",
@@ -1068,8 +1102,7 @@ var generatedTools = [
         },
         EndpointType: {
           type: "string",
-          description: "Request-routing semantics for a modern serverless endpoint. `QUEUE` \u2014 submit asynchronous or synchronous jobs through the managed queue. `LOAD_BALANCER` \u2014 send requests directly to worker-defined HTTP paths.",
-          "x-enum-varnames": ["EndpointTypeQueue", "EndpointTypeLoadBalancer"],
+          description: "Request-routing semantics for a serverless endpoint. `QUEUE` \u2014 submit asynchronous or synchronous jobs through the managed queue. `LOAD_BALANCER` \u2014 send requests directly to worker-defined HTTP paths.",
           enum: ["QUEUE", "LOAD_BALANCER"]
         },
         EndpointWorkers: {
@@ -1178,7 +1211,6 @@ var generatedTools = [
               type: "string",
               minLength: 1,
               maxLength: 191,
-              pattern: "[^\\p{Z}\\t-\\r\x85]",
               description: "Human-readable name, 1\u2013191 characters with at least one non-whitespace character. Leading and trailing whitespace is trimmed.",
               examples: ["training-data"]
             }
@@ -1325,7 +1357,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -1474,7 +1506,7 @@ var generatedTools = [
             {
               type: "object",
               required: ["name"],
-              description: "Request body for creating a pod. Exactly one of `gpu` or `cpu` must be set \u2014 enforced at the handler layer.",
+              description: "Request body for creating a pod. Exactly one of `gpu` or `cpu` must be set. For CPU pods, memory is derived by the API from the selected flavor's RAM multiplier; clients provide only CPU flavor and vCPU count.",
               properties: {
                 name: {
                   type: "string",
@@ -1504,7 +1536,7 @@ var generatedTools = [
                 globalNetworking: {
                   type: "boolean",
                   default: false,
-                  description: "Enable global networking, giving the pod a private IP reachable across data centers. Requires an NVIDIA GPU and a global-networking-enabled data center (both enforced upstream).",
+                  description: "Enable global networking, giving the pod a private IP reachable across data centers. Requires an NVIDIA GPU and a global-networking-enabled data center. See `list-data-centers` (`globalNetwork`) for eligible data centers.",
                   examples: [false]
                 },
                 gpu: {
@@ -1803,7 +1835,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -2264,7 +2296,7 @@ var generatedTools = [
           items: {
             $ref: "#/$defs/CatalogInclude"
           },
-          description: "Comma-separated optional expansions. Supported value today: AVAILABILITY. This may expand with more include values in the future."
+          description: "Comma-separated optional expansions; see `CatalogInclude` for the supported values."
         },
         product: {
           type: "array",
@@ -2284,7 +2316,7 @@ var generatedTools = [
       $defs: {
         CatalogInclude: {
           type: "string",
-          description: "Catalog include expansion. Only AVAILABILITY is supported today; additional include values may be added in the future.",
+          description: "Catalog include expansion. `AVAILABILITY` is the only supported value.",
           enum: ["AVAILABILITY"]
         },
         CpuProduct: {
@@ -2510,7 +2542,7 @@ var generatedTools = [
           items: {
             $ref: "#/$defs/CatalogInclude"
           },
-          description: "Comma-separated optional expansions. Supported value today: AVAILABILITY. This may expand with more include values in the future."
+          description: "Comma-separated optional expansions; see `CatalogInclude` for the supported values."
         },
         product: {
           type: "array",
@@ -2556,7 +2588,7 @@ var generatedTools = [
       $defs: {
         CatalogInclude: {
           type: "string",
-          description: "Catalog include expansion. Only AVAILABILITY is supported today; additional include values may be added in the future.",
+          description: "Catalog include expansion. `AVAILABILITY` is the only supported value.",
           enum: ["AVAILABILITY"]
         },
         GpuCloudFilter: {
@@ -2623,6 +2655,37 @@ var generatedTools = [
       properties: {}
     }
   },
+  {
+    name: "get-organization-user",
+    title: "Get organization user",
+    operationId: "getOrganizationUser",
+    description: "Get an organization member. Returns one member of the authenticated caller's organization by user ID, with the role they hold. Any member can read it. A user who is not a member, including one from another organization, returns 404. Returns 404 when the caller does not belong to an organization, or 403 when organizations are unavailable to the caller.",
+    method: "GET",
+    path: "/v2/organization/users/{id}",
+    params: [
+      {
+        name: "id",
+        location: "path"
+      }
+    ],
+    hasBody: false,
+    annotations: {
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
+    inputSchema: {
+      type: "object",
+      properties: {
+        id: {
+          type: "string",
+          description: "User identifier"
+        }
+      },
+      required: ["id"]
+    }
+  },
   {
     name: "get-pod",
     title: "Get Pod",
@@ -2765,6 +2828,99 @@ var generatedTools = [
       required: ["id"]
     }
   },
+  {
+    name: "list-audit-logs",
+    title: "List audit logs",
+    operationId: "listAuditLogs",
+    description: "List audit logs. Returns the account's audit trail, newest first, cursor-paginated; an omitted `limit` defaults to 200. Each entry records who acted (`actorId`, `email`), what they did (`action`), to which resource (`resourceType`, `resourceId`), and the details of the change (`value`). When the caller belongs to an organization the trail is the organization's, and `actorIds` narrows it to specific members; otherwise it is the trail of the account the request acts for. Filters combine with AND. Entries older than the account's audit log retention window are not returned, and an account whose plan does not include audit logs is answered 403. Follow `pagination.nextCursor` until `hasNextPage` is false.",
+    directoryDescription: "List the account's audit trail, newest first. Results are cursor-paginated and record the actor, action, resource, timestamp, and change details. Filters include actor IDs, actions, time range, and text search. The account's retention window and plan determine which entries are available.",
+    method: "GET",
+    path: "/v2/audit-logs",
+    params: [
+      {
+        name: "actorIds",
+        location: "query",
+        explode: false
+      },
+      {
+        name: "actions",
+        location: "query",
+        explode: false
+      },
+      {
+        name: "startTime",
+        location: "query"
+      },
+      {
+        name: "endTime",
+        location: "query"
+      },
+      {
+        name: "search",
+        location: "query"
+      },
+      {
+        name: "cursor",
+        location: "query"
+      },
+      {
+        name: "limit",
+        location: "query"
+      }
+    ],
+    hasBody: false,
+    annotations: {
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
+    inputSchema: {
+      type: "object",
+      properties: {
+        actorIds: {
+          type: "array",
+          items: {
+            type: "string"
+          },
+          description: "Only return entries performed by these users. Comma-separated user IDs. Available to organization accounts only; a personal or team account is answered 400."
+        },
+        actions: {
+          type: "array",
+          items: {
+            type: "string"
+          },
+          description: "Only return entries whose `action` is one of these names, matched exactly as recorded. Comma-separated."
+        },
+        startTime: {
+          type: "string",
+          format: "date-time",
+          description: "Earliest instant to include, inclusive (RFC 3339). The account's audit log retention window still applies: entries that have aged out are not returned even when this reaches further back."
+        },
+        endTime: {
+          type: "string",
+          format: "date-time",
+          description: "Latest instant to include, inclusive (RFC 3339). Must be after `startTime` when both are given; 400 otherwise."
+        },
+        search: {
+          type: "string",
+          minLength: 1,
+          description: "Case-insensitive substring match against `email`, `resourceType`, `resourceId`, `action` and `value`; `actorId` is not searched. Surrounding whitespace is ignored."
+        },
+        cursor: {
+          type: "string",
+          minLength: 1,
+          description: "Opaque resume cursor \u2014 pass the previous response's `pagination.nextCursor` through verbatim; omit for the first page. A cursor is only valid for the operation and parameters that issued it; a malformed or foreign cursor is rejected."
+        },
+        limit: {
+          type: "integer",
+          minimum: 1,
+          maximum: 200,
+          description: "Page size, 1\u2013200. Defaults to 200 when omitted."
+        }
+      }
+    }
+  },
   {
     name: "list-billing",
     title: "List billing",
@@ -2825,13 +2981,6 @@ var generatedTools = [
         BillingBucketSize: {
           type: "string",
           enum: ["hour", "day", "week", "month", "year"],
-          "x-enum-varnames": [
-            "BillingBucketSizeHour",
-            "BillingBucketSizeDay",
-            "BillingBucketSizeWeek",
-            "BillingBucketSizeMonth",
-            "BillingBucketSizeYear"
-          ],
           default: "day",
           description: "Length of each billing time bucket.",
           examples: ["day"]
@@ -2907,13 +3056,6 @@ var generatedTools = [
         BillingBucketSize: {
           type: "string",
           enum: ["hour", "day", "week", "month", "year"],
-          "x-enum-varnames": [
-            "BillingBucketSizeHour",
-            "BillingBucketSizeDay",
-            "BillingBucketSizeWeek",
-            "BillingBucketSizeMonth",
-            "BillingBucketSizeYear"
-          ],
           default: "day",
           description: "Length of each billing time bucket.",
           examples: ["day"]
@@ -3032,7 +3174,7 @@ var generatedTools = [
           items: {
             $ref: "#/$defs/CatalogInclude"
           },
-          description: "Comma-separated optional expansions. Supported value today: AVAILABILITY. This may expand with more include values in the future."
+          description: "Comma-separated optional expansions; see `CatalogInclude` for the supported values."
         },
         product: {
           type: "array",
@@ -3051,7 +3193,7 @@ var generatedTools = [
       $defs: {
         CatalogInclude: {
           type: "string",
-          description: "Catalog include expansion. Only AVAILABILITY is supported today; additional include values may be added in the future.",
+          description: "Catalog include expansion. `AVAILABILITY` is the only supported value.",
           enum: ["AVAILABILITY"]
         },
         CpuProduct: {
@@ -3270,13 +3412,6 @@ var generatedTools = [
         BillingBucketSize: {
           type: "string",
           enum: ["hour", "day", "week", "month", "year"],
-          "x-enum-varnames": [
-            "BillingBucketSizeHour",
-            "BillingBucketSizeDay",
-            "BillingBucketSizeWeek",
-            "BillingBucketSizeMonth",
-            "BillingBucketSizeYear"
-          ],
           default: "day",
           description: "Length of each billing time bucket.",
           examples: ["day"]
@@ -3513,7 +3648,7 @@ var generatedTools = [
           items: {
             $ref: "#/$defs/CatalogInclude"
           },
-          description: "Comma-separated optional expansions. Supported value today: AVAILABILITY. This may expand with more include values in the future."
+          description: "Comma-separated optional expansions; see `CatalogInclude` for the supported values."
         },
         product: {
           type: "array",
@@ -3558,7 +3693,7 @@ var generatedTools = [
       $defs: {
         CatalogInclude: {
           type: "string",
-          description: "Catalog include expansion. Only AVAILABILITY is supported today; additional include values may be added in the future.",
+          description: "Catalog include expansion. `AVAILABILITY` is the only supported value.",
           enum: ["AVAILABILITY"]
         },
         GpuCloudFilter: {
@@ -3642,13 +3777,6 @@ var generatedTools = [
         BillingBucketSize: {
           type: "string",
           enum: ["hour", "day", "week", "month", "year"],
-          "x-enum-varnames": [
-            "BillingBucketSizeHour",
-            "BillingBucketSizeDay",
-            "BillingBucketSizeWeek",
-            "BillingBucketSizeMonth",
-            "BillingBucketSizeYear"
-          ],
           default: "day",
           description: "Length of each billing time bucket.",
           examples: ["day"]
@@ -3697,6 +3825,26 @@ var generatedTools = [
       }
     }
   },
+  {
+    name: "list-organization-users",
+    title: "List organization users",
+    operationId: "listOrganizationUsers",
+    description: "List organization members. Returns every member of the authenticated caller's organization with the role each holds. Any member can read the roster. The list is complete and takes no pagination parameters. Returns 404 when the caller does not belong to an organization, or 403 when organizations are unavailable to the caller.",
+    method: "GET",
+    path: "/v2/organization/users",
+    params: [],
+    hasBody: false,
+    annotations: {
+      readOnlyHint: true,
+      destructiveHint: false,
+      idempotentHint: true,
+      openWorldHint: true
+    },
+    inputSchema: {
+      type: "object",
+      properties: {}
+    }
+  },
   {
     name: "list-pod-billing",
     title: "List Pod billing",
@@ -3765,13 +3913,6 @@ var generatedTools = [
         BillingBucketSize: {
           type: "string",
           enum: ["hour", "day", "week", "month", "year"],
-          "x-enum-varnames": [
-            "BillingBucketSizeHour",
-            "BillingBucketSizeDay",
-            "BillingBucketSizeWeek",
-            "BillingBucketSizeMonth",
-            "BillingBucketSizeYear"
-          ],
           default: "day",
           description: "Length of each billing time bucket.",
           examples: ["day"]
@@ -3980,13 +4121,6 @@ var generatedTools = [
         BillingBucketSize: {
           type: "string",
           enum: ["hour", "day", "week", "month", "year"],
-          "x-enum-varnames": [
-            "BillingBucketSizeHour",
-            "BillingBucketSizeDay",
-            "BillingBucketSizeWeek",
-            "BillingBucketSizeMonth",
-            "BillingBucketSizeYear"
-          ],
           default: "day",
           description: "Length of each billing time bucket.",
           examples: ["day"]
@@ -4205,7 +4339,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -4302,7 +4436,7 @@ var generatedTools = [
           type: "object",
           required: ["volumeId"],
           additionalProperties: false,
-          description: "A volume mounted on every worker, always at `/runpod-volume`. Custom paths are not supported yet.",
+          description: "A volume mounted on every worker, always at `/runpod-volume`.",
           properties: {
             volumeId: {
               type: "string",
@@ -4562,7 +4696,6 @@ var generatedTools = [
               type: "string",
               minLength: 1,
               maxLength: 191,
-              pattern: "[^\\p{Z}\\t-\\r\x85]",
               description: "New human-readable name, 1\u2013191 characters with at least one non-whitespace character. Leading and trailing whitespace is trimmed.",
               examples: ["renamed-volume"]
             }
@@ -4708,7 +4841,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -4846,7 +4979,7 @@ var generatedTools = [
               properties: {
                 globalNetworking: {
                   type: "boolean",
-                  description: "Enable (true) or disable (false) global networking. Takes effect on the next pod start/restart, not live. Requires an NVIDIA GPU and a global-networking-enabled data center (both enforced upstream)."
+                  description: "Enable (true) or disable (false) global networking. Takes effect on the next pod start/restart, not live. Requires an NVIDIA GPU and a global-networking-enabled data center."
                 },
                 locked: {
                   type: "boolean",
@@ -4993,7 +5126,7 @@ var generatedTools = [
     hasBody: true,
     annotations: {
       readOnlyHint: false,
-      destructiveHint: false,
+      destructiveHint: true,
       idempotentHint: true,
       openWorldHint: true
     },
@@ -5055,7 +5188,7 @@ var generatedTools = [
               description: "Environment variables as key-value pairs",
               examples: [
                 {
-                  JUPYTER_PASSWORD: "hunter2"
+                  JUPYTER_PASSWORD: "changeme"
                 }
               ]
             },
@@ -5175,123 +5308,185 @@ var generatedTools = [
   }
 ];
 
-// src/specgen/profile.ts
-var DIRECTORY_PATH = "/directory";
-var IN = { include: true };
-var CREDENTIAL_INPUT = "Takes a secret or password as input; directories forbid collecting credentials.";
-var ALP = "Collects conversation-derived data beyond the task; the directory profile never serves it.";
-var DIRECTORY_TOOLS = {
-  "cancel-job": IN,
-  "create-cluster": IN,
-  "create-delegation": IN,
-  "create-endpoint": IN,
-  "create-global-volume": IN,
-  "create-network-volume": IN,
-  "create-pod": IN,
-  "create-template": IN,
-  "delete-cluster": IN,
-  "delete-endpoint": IN,
-  "delete-global-volume": IN,
-  "delete-network-volume": IN,
-  "delete-pod": IN,
-  "delete-registry": IN,
-  "delete-secret": IN,
-  "delete-template": IN,
-  "deploy-hub-repo": IN,
-  "endpoint-health": IN,
-  "get-capacity": IN,
-  "get-cluster": IN,
-  "get-cpu-type": IN,
-  "get-data-center": IN,
-  "get-endpoint": IN,
-  "get-endpoint-build": IN,
-  "get-global-volume": IN,
-  "get-gpu-type": IN,
-  "get-job-status": IN,
-  "get-network-volume": IN,
-  "get-organization": IN,
-  "get-pod": IN,
-  "get-registry": IN,
-  "get-secret": IN,
-  "get-ssh-keys": IN,
-  "get-template": IN,
-  "list-billing": IN,
-  "list-cluster-billing": IN,
-  "list-cluster-pods": IN,
-  "list-clusters": IN,
-  "list-cpu-types": IN,
-  "list-data-centers": IN,
-  "list-delegations": IN,
-  "list-endpoint-billing": IN,
-  "list-endpoint-builds": IN,
-  "list-endpoint-releases": IN,
-  "list-endpoint-workers": IN,
-  "list-endpoints": IN,
-  "list-global-volumes": IN,
-  "list-gpu-types": IN,
-  "list-hub-repos": IN,
-  "list-network-volume-billing": IN,
-  "list-network-volumes": IN,
-  "list-pod-billing": IN,
-  "list-pods": IN,
-  "list-public-endpoints": IN,
-  "list-public-templates": IN,
-  "list-registries": IN,
-  "list-secrets": IN,
-  "list-serverless-billing": IN,
-  "list-templates": IN,
-  "lookup-concept": IN,
-  "pod-action": IN,
-  "purge-endpoint-queue": IN,
-  "retry-job": IN,
-  "revoke-delegation": IN,
-  "search-concepts": IN,
-  "search-docs": IN,
-  "set-endpoint-gpus": IN,
-  "stream-job": IN,
-  "stream-pod-logs": IN,
-  "stream-worker-logs": IN,
-  "update-cluster": IN,
-  "update-endpoint": IN,
-  "update-global-volume": IN,
-  "update-network-volume": IN,
-  "update-pod": IN,
-  "update-ssh-keys": IN,
-  "update-template": IN,
-  "run-endpoint": { include: true, omitInputs: ["s3Config"] },
-  "runsync-endpoint": { include: true, omitInputs: ["s3Config"] },
-  "list-guides": {
-    include: false,
-    reason: "Serves playbooks that instruct the model how to behave."
-  },
-  "read-guide": {
-    include: false,
-    reason: "Serves playbooks that instruct the model how to behave."
-  },
-  "describe-tool": {
-    include: false,
-    reason: "Fetches schemas the directory already lists in full; directories reject schema fetching."
-  },
-  "create-secret": { include: false, reason: CREDENTIAL_INPUT },
-  "update-secret": { include: false, reason: CREDENTIAL_INPUT },
-  "create-registry": { include: false, reason: CREDENTIAL_INPUT },
-  report_feedback: { include: false, reason: ALP },
-  report_how_user_found_runpod: { include: false, reason: ALP },
-  save_to_journal: { include: false, reason: ALP },
-  read_journal: { include: false, reason: ALP },
-  ask_question: { include: false, reason: ALP }
+// src/specgen/tools/annotations.ts
+var readOnly = {
+  readOnlyHint: true,
+  destructiveHint: false,
+  idempotentHint: true,
+  openWorldHint: true
 };
-function directoryIncludes(name) {
-  return DIRECTORY_TOOLS[name]?.include === true;
+var bundledRead = {
+  readOnlyHint: true,
+  destructiveHint: false,
+  idempotentHint: true,
+  openWorldHint: false
+};
+var write = {
+  readOnlyHint: false,
+  destructiveHint: false,
+  idempotentHint: false,
+  openWorldHint: true
+};
+var idempotentWrite = {
+  readOnlyHint: false,
+  destructiveHint: false,
+  idempotentHint: true,
+  openWorldHint: true
+};
+var destructive = {
+  readOnlyHint: false,
+  destructiveHint: true,
+  idempotentHint: true,
+  openWorldHint: true
+};
+var destructiveWrite = {
+  readOnlyHint: false,
+  destructiveHint: true,
+  idempotentHint: false,
+  openWorldHint: true
+};
+
+// src/specgen/clients/http-error.ts
+var HttpError = class extends Error {
+  constructor(message, status, payload = void 0) {
+    super(message);
+    this.status = status;
+    this.payload = payload;
+    this.name = "HttpError";
+  }
+};
+function missingKeyError() {
+  return new HttpError(
+    "No Runpod API key: the request carried no usable credential.",
+    401
+  );
+}
+
+// src/specgen/tools/util.ts
+function ok(payload) {
+  return { ok: true, status: 200, payload };
+}
+function badRequest(message) {
+  return { ok: false, status: 400, payload: { error: message } };
+}
+function isTimeout(error) {
+  if (!(error instanceof Error)) return false;
+  if (error.name === "TimeoutError" || error.name === "AbortError") return true;
+  const cause = error.cause;
+  return cause instanceof Error && (cause.name === "TimeoutError" || cause.name === "AbortError");
+}
+var UNREACHED_CODES = /* @__PURE__ */ new Set([
+  "ENOTFOUND",
+  "EAI_AGAIN",
+  "ECONNREFUSED",
+  "EHOSTUNREACH",
+  "ENETUNREACH",
+  "UND_ERR_CONNECT_TIMEOUT"
+]);
+var TLS_CODE = /^(?:CERT_|ERR_TLS_|UNABLE_TO_|DEPTH_ZERO_|SELF_SIGNED_)/;
+function causeCode(cause) {
+  const code = cause?.code;
+  if (typeof code === "string") return code;
+  const inner = cause instanceof AggregateError ? cause.errors[0] : void 0;
+  const innerCode = inner?.code;
+  return typeof innerCode === "string" ? innerCode : void 0;
 }
-function directoryOmittedInputs(name) {
-  const decision = DIRECTORY_TOOLS[name];
-  return decision?.include ? decision.omitInputs ?? [] : [];
+function networkFailureResult(error) {
+  if (!(error instanceof TypeError)) return void 0;
+  const afterHeaders = error.message === "terminated";
+  if (!afterHeaders && error.message !== "fetch failed") return void 0;
+  const code = causeCode(error.cause);
+  if (!code) return void 0;
+  const unreached = !afterHeaders && (UNREACHED_CODES.has(code) || TLS_CODE.test(code));
+  return {
+    ok: false,
+    status: unreached ? 503 : 502,
+    requestDelivery: unreached ? "not_sent" : "uncertain",
+    payload: unreached ? {
+      error: "Could not reach the Runpod API.",
+      cause: code,
+      hint: "The request never reached Runpod, so retrying the same call is safe: retry it once. If it fails again, the network, DNS, proxy or TLS setup on this machine is the likely cause; check https://uptime.runpod.io and report it rather than retrying further."
+    } : {
+      error: "The connection to the Runpod API dropped before the full response arrived.",
+      cause: code,
+      hint: "The request may have been applied. A read can be retried once. For a create, update, delete or job submission, check with the matching get- or list- tool whether it took effect before calling again; retrying blindly can create a duplicate billable resource."
+    }
+  };
 }
-function profileForPath(pathname) {
-  return pathname.replace(/\/+$/, "") === DIRECTORY_PATH ? "directory" : "default";
+async function runTool(body) {
+  try {
+    return await body();
+  } catch (error) {
+    if (error instanceof HttpError) {
+      return {
+        ok: false,
+        status: error.status,
+        payload: { error: error.message, detail: error.payload }
+      };
+    }
+    if (isTimeout(error)) {
+      return {
+        ok: false,
+        status: 504,
+        requestDelivery: "uncertain",
+        payload: {
+          error: "The Runpod API did not respond before the request deadline.",
+          hint: "Transient upstream stall: retry the same call once; if it persists, check https://uptime.runpod.io."
+        }
+      };
+    }
+    const network = networkFailureResult(error);
+    if (network) return network;
+    throw error;
+  }
 }
 
+// src/specgen/tools/pod-transitions.ts
+var podAction = generatedTools.find((tool) => tool.name === "pod-action");
+if (!podAction)
+  throw new Error("pod-action is missing from the generated tools");
+var POD_ID = {
+  type: "object",
+  properties: { id: { type: "string", description: "ID of the Pod" } },
+  required: ["id"],
+  additionalProperties: false
+};
+function transition(action, title, description, annotations) {
+  return {
+    name: `${action}-pod`,
+    title,
+    description,
+    inputSchema: POD_ID,
+    annotations,
+    handler: (ctx, args) => runTool(
+      () => dispatchGeneratedTool(ctx.sdk, podAction, {
+        id: args.id,
+        body: { action }
+      })
+    )
+  };
+}
+var podTransitionTools = [
+  transition(
+    "start",
+    "Start Pod",
+    "Start a stopped Pod (status EXITED or ERROR). Billable: the Pod accrues its hourly price while it runs. Returns the updated Pod; a Pod in any other status returns 409.",
+    write
+  ),
+  transition(
+    "stop",
+    "Stop Pod",
+    "Stop a running or provisioning Pod. It releases its GPU or CPU and stops compute billing; the container disk is wiped, while the volume disk and any attached network volume keep their data and volume storage keeps billing. Returns the updated Pod; a Pod in any other status returns 409.",
+    destructiveWrite
+  ),
+  transition(
+    "restart",
+    "Restart Pod",
+    "Restart a running Pod's container in place. Whatever is running in the container is interrupted and the container disk is wiped; the volume disk and any attached network volume keep their data. Returns the updated Pod; a Pod that is not RUNNING returns 409.",
+    destructiveWrite
+  )
+];
+
 // src/specgen/redact.ts
 var REDACTED = "[redacted]";
 var SECRET_KEY = /pass(word|wd)?|secret|token|api_?key|private_?key|credential|auth/i;
@@ -5390,137 +5585,6 @@ function capList(items, options, extra) {
   };
 }
 
-// src/specgen/clients/http-error.ts
-var HttpError = class extends Error {
-  constructor(message, status, payload = void 0) {
-    super(message);
-    this.status = status;
-    this.payload = payload;
-    this.name = "HttpError";
-  }
-};
-function missingKeyError() {
-  return new HttpError(
-    "No Runpod API key: the request carried no usable credential.",
-    401
-  );
-}
-
-// src/specgen/tools/util.ts
-function ok(payload) {
-  return { ok: true, status: 200, payload };
-}
-function badRequest(message) {
-  return { ok: false, status: 400, payload: { error: message } };
-}
-function isTimeout(error) {
-  if (!(error instanceof Error)) return false;
-  if (error.name === "TimeoutError" || error.name === "AbortError") return true;
-  const cause = error.cause;
-  return cause instanceof Error && (cause.name === "TimeoutError" || cause.name === "AbortError");
-}
-var UNREACHED_CODES = /* @__PURE__ */ new Set([
-  "ENOTFOUND",
-  "EAI_AGAIN",
-  "ECONNREFUSED",
-  "EHOSTUNREACH",
-  "ENETUNREACH",
-  "UND_ERR_CONNECT_TIMEOUT"
-]);
-var TLS_CODE = /^(?:CERT_|ERR_TLS_|UNABLE_TO_|DEPTH_ZERO_|SELF_SIGNED_)/;
-function causeCode(cause) {
-  const code = cause?.code;
-  if (typeof code === "string") return code;
-  const inner = cause instanceof AggregateError ? cause.errors[0] : void 0;
-  const innerCode = inner?.code;
-  return typeof innerCode === "string" ? innerCode : void 0;
-}
-function networkFailureResult(error) {
-  if (!(error instanceof TypeError)) return void 0;
-  const afterHeaders = error.message === "terminated";
-  if (!afterHeaders && error.message !== "fetch failed") return void 0;
-  const code = causeCode(error.cause);
-  if (!code) return void 0;
-  const unreached = !afterHeaders && (UNREACHED_CODES.has(code) || TLS_CODE.test(code));
-  return {
-    ok: false,
-    status: unreached ? 503 : 502,
-    payload: unreached ? {
-      error: "Could not reach the Runpod API.",
-      cause: code,
-      hint: "The request never reached Runpod, so retrying the same call is safe: retry it once. If it fails again, the network, DNS, proxy or TLS setup on this machine is the likely cause; check https://uptime.runpod.io and report it rather than retrying further."
-    } : {
-      error: "The connection to the Runpod API dropped before the full response arrived.",
-      cause: code,
-      hint: "The request may have been applied. A read can be retried once. For a create, update, delete or job submission, check with the matching get- or list- tool whether it took effect before calling again; retrying blindly can create a duplicate billable resource."
-    }
-  };
-}
-async function runTool(body) {
-  try {
-    return await body();
-  } catch (error) {
-    if (error instanceof HttpError) {
-      return {
-        ok: false,
-        status: error.status,
-        payload: { error: error.message, detail: error.payload }
-      };
-    }
-    if (isTimeout(error)) {
-      return {
-        ok: false,
-        status: 504,
-        payload: {
-          error: "The Runpod API did not respond before the request deadline.",
-          hint: "Transient upstream stall: retry the same call once; if it persists, check https://uptime.runpod.io."
-        }
-      };
-    }
-    const network = networkFailureResult(error);
-    if (network) return network;
-    throw error;
-  }
-}
-
-// src/specgen/tools/annotations.ts
-var readOnly = {
-  readOnlyHint: true,
-  destructiveHint: false,
-  idempotentHint: true,
-  openWorldHint: true
-};
-var bundledRead = {
-  readOnlyHint: true,
-  destructiveHint: false,
-  idempotentHint: true,
-  openWorldHint: false
-};
-var write = {
-  readOnlyHint: false,
-  destructiveHint: false,
-  idempotentHint: false,
-  openWorldHint: true
-};
-var idempotentWrite = {
-  readOnlyHint: false,
-  destructiveHint: false,
-  idempotentHint: true,
-  openWorldHint: true
-};
-var destructive = {
-  readOnlyHint: false,
-  destructiveHint: true,
-  idempotentHint: true,
-  openWorldHint: true
-};
-var destructiveWrite = {
-  readOnlyHint: false,
-  destructiveHint: true,
-  idempotentHint: false,
-  openWorldHint: true
-};
-
 // src/specgen/tools/capacity.ts
 var CUDA_VERSION_REGEX = /^\d{1,2}\.\d{1,2}$/;
 var MAX_PROBE_VERSIONS = 12;
@@ -6903,7 +6967,7 @@ function knowledgeSearchBase(env) {
   return pick(env.KNOWLEDGE_SEARCH_URL, "") || void 0;
 }
 
-// node_modules/.pnpm/@runpod+typescript-api-sdk@0.3.0/node_modules/@runpod/typescript-api-sdk/dist/index.js
+// packages/typescript-api-sdk/dist/index.js
 var PATH_PARAM_RE = /\{[^{}]+\}/g;
 var supportsRequestInitExt = () => {
   return typeof process === "object" && Number.parseInt(process?.versions?.node?.substring(0, 2)) >= 18 && process.versions.undici;
@@ -7475,10 +7539,7 @@ var Parser = class {
   parseItem(standaloneItem = true) {
     if (standaloneItem)
       this.skipWS();
-    const result = [
-      this.parseBareItem(),
-      this.parseParameters()
-    ];
+    const result = [this.parseBareItem(), this.parseParameters()];
     if (standaloneItem)
       this.checkTrail();
     return result;
@@ -7498,10 +7559,7 @@ var Parser = class {
       this.skipWS();
       if (this.lookChar() === ")") {
         this.pos++;
-        return [
-          innerList,
-          this.parseParameters()
-        ];
+        return [innerList, this.parseParameters()];
       }
       innerList.push(this.parseItem(false));
       const nextChar = this.lookChar();
@@ -8088,7 +8146,6 @@ var LogEventParser = class {
   constructor(maxEventBytes) {
     this.maxEventBytes = maxEventBytes;
   }
-  maxEventBytes;
   line = "";
   dataLines = [];
   eventName = "";
@@ -8834,7 +8891,9 @@ function listTool(tool, profile = "default") {
         tool.directoryDescription ?? tool.description,
         inputSchema
       ),
-      inputSchema: withoutInputs(inputSchema, directoryOmittedInputs(name)),
+      inputSchema: withoutCredentialExamples(
+        withoutInputs(inputSchema, directoryOmittedInputs(name))
+      ),
       annotations
     };
   const { description } = tool;
@@ -8856,6 +8915,21 @@ function withApiReference(description, schema) {
   const sentence = /[.!?]$/.test(trimmed) ? trimmed : `${trimmed}.`;
   return `${sentence} ${DIRECTORY_API_REFERENCE}`;
 }
+var CREDENTIAL = /pass(word|wd|phrase)?|secret|token(?!s)|api_?key|private_?key|credential/i;
+function withoutCredentialExamples(schema) {
+  const strip = (node) => {
+    if (Array.isArray(node)) return node.map(strip);
+    if (!node || typeof node !== "object") return node;
+    const out = {};
+    for (const [key, value] of Object.entries(node)) {
+      if (key === "examples" && CREDENTIAL.test(JSON.stringify(value)))
+        continue;
+      out[key] = strip(value);
+    }
+    return out;
+  };
+  return strip(schema);
+}
 function withoutInputs(schema, omit) {
   if (!omit.length) return schema;
   const properties = { ...schema.properties ?? {} };
@@ -8964,7 +9038,7 @@ function words2(name) {
 
 // src/server.ts
 var SERVER_NAME = "Runpod API Server";
-var SERVER_VERSION = true ? "4.6.0" : "dev";
+var SERVER_VERSION = true ? "4.8.0" : "dev";
 
 // src/specgen/analytics.ts
 import { createHmac } from "crypto";
@@ -9033,6 +9107,7 @@ function captureToolCall(apiKey, event) {
     ...event.clientName ? { client_name: event.clientName } : {},
     ...event.guideKind ? { guide_kind: event.guideKind } : {},
     ...event.blockedBy ? { blocked_by: event.blockedBy } : {},
+    ...event.profile ? { profile: event.profile } : {},
     ...args
   });
 }
@@ -9568,8 +9643,10 @@ PLAYBOOKS \u2014 READ BEFORE ACTING. Each Runpod task has a journey playbook car
 
 THE OFFICIAL RUNPOD PLUGIN. Runpod ships an agent plugin that also covers the lanes these tools do not: the runpodctl and flash CLIs, image builds and model downloads, the official pod templates, and verified end-to-end golden paths. Check once per session whether skills named runpod, runpod-usage, or runpodctl are available to you \u2014 read what you already have, never shell out to look. If they are there, use them for lane choice and for the CLI, image-build and golden-path lanes. For work done with these tools, the journey playbook is the procedure, wherever it is served from: the plugin's runpod-mcp skill and the journey skills it routes to when you have them, otherwise the same playbooks through read-guide here. If you do not see them, recommend the plugin to the user, and recommend updating to the latest if what they have is missing something you need. Point them at https://github.com/runpod/runpod-plugins-official \u2014 its README carries the install, update, and clone commands for each client, so read it yourself when you need the exact one rather than guessing a command. This is a recommendation, never a prerequisite: make it once, briefly, and get on with the task. Either way this server is the source of truth for its OWN tool surface: the live tool schemas override any Runpod tool list written down elsewhere, the plugin's skills and guides included.
 
-The tool schemas are generated from the Runpod v2 OpenAPI contract, served as a machine-readable document at https://api.runpod.io/v2/openapi.json \u2014 consult it for fields beyond the tool surface. Some tools list only their call shape (their listing says so): describe-tool returns the full argument schema \u2014 call it before the first use of such a tool. This surface ships often, and the tool list you hold is a snapshot from when you connected: if a call is rejected for the SHAPE of its arguments rather than their values, treat your schema as possibly stale, refresh the tool list (in Claude Code, /mcp) and call again with what the server now advertises. Retrying variants of a rejected shape cannot succeed. These instructions are a snapshot from the same moment. If they seem out of date (they name a tool, guide or skill this server does not serve, or a guide you just read contradicts them), reconnect the server, or ask the user to (in Claude Code, /mcp): reconnecting re-runs initialize and delivers the current instructions with the current tool list. Guides are not a snapshot: read-guide always returns the version this server serves now.`;
-var DIRECTORY_INSTRUCTIONS = `These tools manage a Runpod account: the compute catalog (GPU and CPU types, data centers, live stock and pricing), Pods, Serverless endpoints and their jobs, Instant Clusters, templates, network and global volumes, secrets, SSH keys, container registry credentials, billing history, Public Endpoints, and the Runpod Hub. Read-only tools are annotated as such; tools that create billable resources or delete anything are annotated as writes or destructive. Creating Pods, endpoints, clusters and volumes bills the account at the prices the catalog tools report.`;
+REST V2 REFERENCE. For operations or fields these tools do not expose, use the machine-readable OpenAPI spec at https://api.runpod.io/v2/openapi.json or the API reference at https://docs.runpod.io/api-reference-v2/overview. Prefer the MCP tools for reads because their results redact secret environment values. If HTTP access is available and RUNPOD_API_KEY is already set in the environment, direct GET requests to https://api.runpod.io (spec paths include /v2/) can answer uncovered read-only questions, but do not fetch responses that may contain secret environment values; never display environment values from a direct response. Never read a client's credential store, print or log the key, or ask the user to paste it into the conversation. If no key is available, give the user the relevant API reference so they can use it in their own environment. Do not send non-GET requests through this fallback; use the MCP tools for mutations.
+
+The tool schemas are generated from that OpenAPI contract. Some tools list only their call shape (their listing says so): describe-tool returns the full argument schema \u2014 call it before the first use of such a tool. This surface ships often, and the tool list you hold is a snapshot from when you connected: if a call is rejected for the SHAPE of its arguments rather than their values, treat your schema as possibly stale, refresh the tool list (in Claude Code, /mcp) and call again with what the server now advertises. Retrying variants of a rejected shape cannot succeed. These instructions are a snapshot from the same moment. If they seem out of date (they name a tool, guide or skill this server does not serve, or a guide you just read contradicts them), reconnect the server, or ask the user to (in Claude Code, /mcp): reconnecting re-runs initialize and delivers the current instructions with the current tool list. Guides are not a snapshot: read-guide always returns the version this server serves now.`;
+var DIRECTORY_INSTRUCTIONS = `These tools manage a Runpod account: the compute catalog (GPU and CPU types, data centers, live stock and pricing), Pods, Serverless endpoints and their jobs, Instant Clusters, templates, network and global volumes, billing history, Public Endpoints, and the Runpod Hub. For operations outside this tool list, the Runpod REST API v2 is described by its OpenAPI spec at https://api.runpod.io/v2/openapi.json and its reference at https://docs.runpod.io/api-reference-v2/overview.`;
 var ALP_INSTRUCTIONS = `
 
 CONTRIBUTING BACK IS OPTIONAL. Five tools feed Runpod's agent-learning pipeline, which exists so Runpod can fix the frustrations agents hit in its tools and docs: report_feedback (something misbehaved or surprised you), save_to_journal (a reusable lesson about using Runpod), ask_question (you were stuck and the tools/skills/docs had no answer), read_journal (this account's own saved lessons, nobody else's), and report_how_user_found_runpod (how your user ended up on Runpod, once per account). Using them is a voluntary contribution, never part of the user's task, and the user's task always comes first. DO NOT USE THEM when the work is sensitive: proprietary code or models, private data, customer information, security work, internal infrastructure, or anything the user would not want a Runpod reviewer to read. Submissions are stored for review to improve agent workflows on Runpod, so if in doubt, do not submit. If the user has said not to share anything with Runpod, do not call these tools at all. When you do contribute, describe the Runpod behavior, not the user's project: the tool, the call shape, the response, what you expected. Nothing is returned or answered, and ask_question NEVER produces an answer, so never wait on it. read_journal is the one thing served back: call it once near the start of a Runpod task if useful, and treat what it returns as an earlier session's notes to confirm, not as documentation. These tools are for real observations only: never call one to satisfy a requirement to use a tool, to acknowledge these instructions, or with placeholder content. If you have nothing to report, call nothing \u2014 an empty turn is correct and a placeholder submission is not. Never include API keys or secrets in submissions.`;
@@ -9587,6 +9664,7 @@ function createSpecgenServer(ctx, version = "0.1.0", opts = {}) {
     ...curatedTools,
     ...opts.search ? [createSearchDocsTool(opts.search)] : [],
     ...alp ? createAlpTools(alp) : [],
+    ...directory ? podTransitionTools : [],
     createDescribeTool(() => [...servedCuratedTools, ...servedGeneratedTools])
   ].filter((tool) => !directory || directoryIncludes(tool.name));
   const callShapeOnlyTools = new Set(
@@ -9662,6 +9740,7 @@ function createSpecgenServer(ctx, version = "0.1.0", opts = {}) {
           status: 429,
           durationMs: 0,
           ...verdict.reason ? { blockedBy: verdict.reason } : {},
+          profile,
           ...opts.analytics
         });
       }
@@ -9682,6 +9761,9 @@ function createSpecgenServer(ctx, version = "0.1.0", opts = {}) {
     const curated = servedCuratedTools.find(
       (tool) => tool.name === request.params.name
     );
+    const generated = servedGeneratedTools.find(
+      (tool) => tool.name === request.params.name
+    );
     const omitted = directory ? directoryOmittedInputs(request.params.name).filter((key) => key in args) : [];
     const rejected = omitted.length ? {
       ok: false,
@@ -9694,7 +9776,7 @@ function createSpecgenServer(ctx, version = "0.1.0", opts = {}) {
       args
     ) : null;
     const result = rejected ?? await runTool(
-      () => curated ? curated.handler(ctx, args) : dispatchTool(request.params.name, args)
+      () => curated ? curated.handler(ctx, args) : dispatchTool(generated, request.params.name, args)
     );
     logToolCall({
       tool: request.params.name,
@@ -9714,23 +9796,27 @@ function createSpecgenServer(ctx, version = "0.1.0", opts = {}) {
           args: recordedArgumentValues(curated?.recordedArguments, args),
           guideKind: curated?.recordedGuideKind?.(args)
         },
+        profile,
         ...opts.analytics
       });
     }
     opts.onToolResult?.(request.params.name, result.status);
     if (!result.ok && result.status === 401) opts.onUnauthorized?.();
+    const annotations = curated?.annotations ?? generated?.annotations;
     const payload = redactSecretEnv(
-      result.ok ? result.payload : withRecoveryHint(result, callShapeOnlyTools.has(request.params.name))
+      result.ok ? result.payload : withRecoveryHint(
+        result,
+        annotations?.idempotentHint === true,
+        request.params.name,
+        callShapeOnlyTools.has(request.params.name)
+      )
     );
     return {
       content: [{ type: "text", text: JSON.stringify(payload, null, 2) }],
       isError: !result.ok
     };
   });
-  async function dispatchTool(name, args) {
-    const tool = servedGeneratedTools.find(
-      (candidate) => candidate.name === name
-    );
+  async function dispatchTool(tool, name, args) {
     if (!tool)
       return {
         ok: false,
@@ -9778,8 +9864,24 @@ function errorHint(status, detail) {
   return void 0;
 }
 var CALL_SHAPE_ONLY_HINT = `This tool lists its call shape only; ${DESCRIBE_TOOL_NAME} serves the full argument schema.`;
-function withRecoveryHint(result, listsCallShapeOnly = false) {
+var OUTCOME_UNKNOWN_HINT = "Runpod may have completed this operation even though its response failed. Do not submit it again yet. Check the matching get- or list- tool or the Runpod console using the original request details and time. Lists can lag; if you cannot confirm the outcome, stop and ask the user to verify before retrying.";
+var JOB_OUTCOME_UNKNOWN_HINT = "Runpod may have submitted this Serverless job even though its response failed. Do not submit it again yet. No job-list tool is available, and get-job-status requires a job ID that this failed response may not have returned. Check the endpoint in the Runpod console or ask the user to verify the submission. If the outcome remains uncertain, stop instead of retrying.";
+var RETRY_JOB_OUTCOME_UNKNOWN_HINT = "Runpod may have requeued this Serverless job even though its response failed. Do not retry it again yet. Check get-job-status with the endpoint and job IDs you supplied, or check the Runpod console. If you cannot confirm the outcome, stop and ask the user to verify before retrying.";
+var JOB_SUBMISSION_TOOLS = /* @__PURE__ */ new Set(["run-endpoint", "runsync-endpoint"]);
+function withRecoveryHint(result, idempotent, toolName, listsCallShapeOnly = false) {
   const payload = asRecord(result.payload) ?? { error: result.payload };
+  if (!idempotent && result.status >= 500 && result.requestDelivery !== "not_sent") {
+    const detail = asRecord(payload.detail);
+    const safeDetail = detail ? { ...detail } : void 0;
+    if (safeDetail) delete safeDetail.hint;
+    return {
+      ...payload,
+      ...safeDetail ? { detail: safeDetail } : {},
+      ...payload.code !== void 0 ? { upstreamCode: payload.code } : {},
+      code: "outcome_unknown",
+      hint: toolName === "retry-job" ? RETRY_JOB_OUTCOME_UNKNOWN_HINT : JOB_SUBMISSION_TOOLS.has(toolName) ? JOB_OUTCOME_UNKNOWN_HINT : OUTCOME_UNKNOWN_HINT
+    };
+  }
   const hint = payload.hint ?? asRecord(payload.detail)?.hint ?? errorHint(result.status, detailText(payload));
   const shapeRejected = result.status === 400 || result.status === 422;
   return {