@writepanda/mcp 1.234.0 → 1.241.0

package/bin/server.mjs

@@ -3065,6 +3065,58 @@ const TOOLS = [
 		},
 		command: "audio.probe",
 	},
+	{
+		name: "audio_detect_beats",
+		description:
+			"Beats of a music track on the EDITED timeline: beatsMs, downbeatsMs (bar starts, 4/4 estimated), tempoBpm, confidence. Local analysis, cached per file; follows the track's trim, in-point and loop. regionId = audio overlay id (default: the longest music track). Use to time cuts, graphics or B-roll to music; project_place_on_beats does a whole montage. No beats / confidence < 0.2 = no steady beat.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				regionId: {
+					type: "string",
+					description: "Audio overlay id. Default: the longest music track.",
+				},
+				fromMs: { type: "number", minimum: 0, description: "Only beats from here (edited ms)." },
+				toMs: { type: "number", minimum: 0, description: "Only beats up to here (edited ms)." },
+				force: { type: "boolean", description: "Re-analyse instead of using the cache." },
+			},
+		},
+		command: "audio.detect-beats",
+	},
+	{
+		name: "project_place_on_beats",
+		description:
+			"Music-driven montage: lay pictures / B-roll / video overlays out so each changes on the beat. regionIds = existing overlay ids in play order (retimed; B-roll transitions follow) OR paths = new image/video files to add full-frame. every: 1 (every beat), 2 (default), 4 (every bar, from a downbeat). fromMs/toMs bound the run (default: the music's span). Item i plays from cut i to cut i+1; extras come back in unplaced.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				regionIds: {
+					type: "array",
+					items: { type: "string" },
+					description: "Overlay ids, in play order.",
+				},
+				paths: {
+					type: "array",
+					items: { type: "string" },
+					description: "Or image/video files to add, in play order.",
+				},
+				every: { type: "number", enum: [1, 2, 4], description: "Beats per item. Default 2." },
+				fromMs: { type: "number", minimum: 0 },
+				toMs: { type: "number", minimum: 0 },
+				musicRegionId: {
+					type: "string",
+					description: "Audio overlay to follow. Default: the longest music track.",
+				},
+				expectedRevision: { type: "number" },
+			},
+			required: ["id"],
+		},
+		command: "project.place-on-beats",
+	},
 	{
 		name: "project_set_region_sound",
 		description: "Set or clear the SFX on an existing zoom, motionGraphic or fx region.",
@@ -3952,9 +4004,14 @@ const TOOLS = [
 				fadeOut: { type: "number", description: "ms; needs endMs." },
 				anchorSourceMs: { type: "number", description: "SFX only (not music): SOURCE ms anchor." },
 				anchorSourceEndMs: ANCHOR_END_PROP,
+				loop: {
+					type: "boolean",
+					description: "Repeat the file to endMs (a ~30 s Lyria clip under a longer video).",
+				},
 				transcribe: {
 					type: "boolean",
-					description: "Voiceover: merge its words into the transcript (job_wait transcribeJobId).",
+					description:
+						"Words into the transcript (job_wait transcribeJobId). A generated narration is in by DEFAULT (script timed to the audio); false keeps it out. true transcribes any other voiceover.",
 				},
 				ducking: {
 					description: "Duck under the voice: true or a ducking object.",
@@ -4069,6 +4126,46 @@ const TOOLS = [
 		},
 		command: "project.set-clip-color",
 	},
+	{
+		name: "project_align_cursor",
+		description:
+			"Fix a screen recording whose drawn cursor / cursor auto-zooms land in the wrong place (old recordings saved the pointer with the wrong mapping) without re-recording. Pins: at atMs the REAL pointer (or what was clicked) is at x,y, fractions 0-1 of the recorded frame. 1 pin = offset; 2+ pins far apart also fix scale per axis. Stored on the clip (non-destructive), used by preview, export and auto-zoom; cursor auto-zooms follow unless moveZooms false. reset removes it. Check with project_render_frame.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				clipId: {
+					type: "string",
+					description: "Default: the clip under the first pin (or the only clip).",
+				},
+				points: {
+					type: "array",
+					description:
+						"[{atMs (edited timeline ms) or sourceMs (clip's own ms), x 0-1, y 0-1}] where the real pointer is.",
+					items: {
+						type: "object",
+						properties: {
+							atMs: { type: "number" },
+							sourceMs: { type: "number" },
+							x: { type: "number" },
+							y: { type: "number" },
+						},
+						required: ["x", "y"],
+					},
+				},
+				add: { type: "boolean", description: "Add to the clip's existing pins (refine)." },
+				reset: { type: "boolean", description: "Remove the correction." },
+				moveZooms: {
+					type: "boolean",
+					description: "Re-point cursor-placed auto-zooms with it (default true).",
+				},
+				dryRun: { type: "boolean" },
+				expectedRevision: { type: "number" },
+			},
+		},
+		command: "project.align-cursor",
+	},
 	{
 		name: "project_set_clip_chroma_key",
 		description:
@@ -4402,7 +4499,7 @@ const TOOLS = [
 	{
 		name: "transcript_transcribe",
 		description:
-			"Transcribe clips with word timestamps (default: only clips without words). Engine: language auto/english = Parakeet; any other named language = local Whisper LOCKED to it (engine override). Returns { jobId, plan } (plan = engine + reason, ranges, totalAudioMs: the audio that will be processed; tell the user for long jobs). To REDO a transcript (wrong language, bad run) confirm language, pass per-run language and provider overrides, then force: true (all clips) or clipId. Word fixes are re-applied; tell the user about result droppedWordEdits[] (fixes that may need redoing). Each clip is transcribed only over the part it plays (head trim to tail trim, e.g. one half of a split; only the kept parts when cuts hide most of a long recording; result transcribedRanges[]). ASYNC: { jobId }, job_wait. Concurrent saves are merged and retried (three attempts). Exhausted conflicts report code revision_conflict / status 409 (job.errorDetails for async jobs).",
+			"Transcribe clips with word timestamps (default: only clips without words). Engine: language auto/english = Parakeet; any other named language = local Whisper LOCKED to it (engine override). Returns { jobId, plan } (plan = engine + reason, ranges, totalAudioMs: the audio that will be processed, estimate/estimatedMs: how long on this computer, advice: faster options for a long Whisper run; tell the user for long jobs). Long Whisper runs are saved chunk by chunk: stalls restart from the last chunk automatically; after job_cancel, a crash or an app restart, transcribing again resumes there. Progress is real (audio decoded); time left in progress.message and progress.data.etaMs. To REDO a transcript (wrong language, bad run) confirm language, pass per-run language and provider overrides, then force: true (all clips) or clipId. Word fixes are re-applied; tell the user about result droppedWordEdits[] (fixes that may need redoing). Each clip is transcribed only over the part it plays (head trim to tail trim, e.g. one half of a split; only the kept parts when cuts hide most of a long recording; result transcribedRanges[]). ASYNC: { jobId }, job_wait. Concurrent saves are merged and retried (three attempts). Exhausted conflicts report code revision_conflict / status 409 (job.errorDetails for async jobs).",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -4443,6 +4540,26 @@ const TOOLS = [
 		inputSchema: { type: "object", properties: {} },
 		command: "system.download-whisper-model",
 	},
+	{
+		name: "transcript_align_narration",
+		description:
+			"Generated narrations into the transcript and captions (script timed to the audio with local speech recognition: the words are exactly the script). New narrations are aligned automatically; use for older projects (default: every narration with a script but no words), force re-timing, or include false to keep one out (sung / non-speech) / true to put it back. ASYNC { jobIds }: job_wait each.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				regionId: {
+					type: "string",
+					description: "One narration (audio-N); omit for all that need it.",
+				},
+				include: { type: "boolean", description: "false = out of transcript + captions." },
+				force: { type: "boolean", description: "Re-align even with words." },
+				expectedRevision: { type: "number" },
+			},
+		},
+		command: "transcript.align-narration",
+	},
 	{
 		name: "transcript_import",
 		description:
@@ -4665,7 +4782,7 @@ const TOOLS = [
 	{
 		name: "transcript_remove_silences",
 		description:
-			"Remove measured quiet runs inside transcript pauses (never into a word) with a 50ms decay hold. Counts only visible time (nothing under a head trim or existing cut). Returns snapped/skipped/warnings; snapToAudio=false uses legacy timing.",
+			"Remove measured quiet runs inside transcript pauses (never into a word) with a 50ms decay hold. Counts only visible time (nothing under a head trim or existing cut). Returns snapped/skipped/warnings; snapToAudio=false uses legacy timing. Applies its cuts to the project as it is when it writes (editor saves meanwhile are kept, no revision_conflict). Fails with code media_offline / media_read_stalled / media_read_too_slow when the source (e.g. an online-only Dropbox / iCloud file) can't be read: tell the user to make it available offline, then run again; project_changed_during_analysis when the clips or word timings kept changing.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -5056,7 +5173,7 @@ const TOOLS = [
 	{
 		name: "media_regenerate_narration",
 		description:
-			"Regenerate a narration already on the timeline IN PLACE from its stored source (script, model, voice, tone on audioOverlays[].generatedFrom). Pass only what changes (text with the edited script, voice, model, ...). Same start, volume, fades, ducking; length = new audio. A transcribed narration is re-transcribed (transcribeJobId: job_wait it). One undo step. No stored source (older narration) = pass text.",
+			"Regenerate a narration already on the timeline IN PLACE from its stored source (script, model, voice, tone on audioOverlays[].generatedFrom). Pass only what changes (text with the edited script, voice, model, ...). Same start, volume, fades, ducking; length = new audio. Its words in the transcript are re-timed to the new take from the script (transcribeJobId: job_wait it) unless it opted out. One undo step. No stored source (older narration) = pass text.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -5585,30 +5702,38 @@ const TOOLS = [
 	{
 		name: "media_generate_music",
 		description:
-			"Generate instrumental music via Replicate: lyria (default, ~30 s) or musicgen (durationSec, can match a reference the user has rights to). Returns { audioPath, durationMs }.",
+			"Generate a ~30 s instrumental music clip with Google Lyria 2 via Replicate (paid, user's account). Under a longer video add it with project_add_audio loop=true endMs=<video>. Returns { audioPath, durationMs, generatedFrom }. MusicGen was removed (non-commercial licence): no model choice, reference, continuation or length.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				prompt: { type: "string", description: "Genre, mood, instruments, tempo." },
-				negativePrompt: { type: "string", description: "lyria: qualities to avoid." },
+				negativePrompt: { type: "string", description: "Qualities to avoid." },
 				seed: { type: "number" },
-				model: { type: "string", enum: ["lyria", "musicgen"] },
-				reference: {
-					type: "string",
-					description: "musicgen: absolute path (audio or video) to match.",
-				},
-				referenceStartMs: { type: "number" },
-				referenceDurationMs: { type: "number", description: "Default 15000, max 30000." },
-				continuation: { type: "boolean" },
-				durationSec: {
-					type: "number",
-					description: "musicgen: default 30, max 300.",
-				},
 			},
 			required: ["prompt"],
 		},
 		command: "media.generate-music",
 	},
+	{
+		name: "media_regenerate_music",
+		description:
+			"Regenerate a generated music track on the timeline IN PLACE with Lyria from its stored source (prompt, avoid, seed on audioOverlays[].generatedFrom). Pass only what changes; newSeed for a fresh variation. Keeps start, level, fades, ducking; a looping bed keeps its window. A track made with MusicGen (removed) regenerates with Lyria. Paid Replicate run, only when asked.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				regionId: { type: "string", description: "The track's audio overlay id (audio-N)." },
+				prompt: { type: "string" },
+				negativePrompt: { type: "string", description: "Avoid; '' clears." },
+				seed: { type: "number" },
+				newSeed: { type: "boolean", description: "A fresh variation." },
+				expectedRevision: { type: "number" },
+			},
+			required: ["regionId"],
+		},
+		command: "media.regenerate-music",
+	},
 	{
 		name: "media_remove_background",
 		description:
@@ -6438,7 +6563,7 @@ const TOOLS = [
 	{
 		name: "clips_make",
 		description:
-			"Long video (podcast, webinar, talk) in, scored vertical clips out, no agent needed: imports, transcribes, finds the best sentence-aligned moments across the whole video (local AI + transcript score), and forks each into its own 9:16 project with captions and face-following framing. Async: returns { clipSetId, jobId }; poll clips_get. Then clips_export, or edit a clip's projectPath with any project tool.",
+			"Long video (podcast, webinar, talk) in, scored vertical clips out, no agent needed: imports, transcribes, finds the best sentence-aligned moments across the whole video (local AI + transcript score), and forks each into its own 9:16 project with captions and face-following framing. Async: returns { clipSetId, jobId, transcription: { engine, estimate, estimatedMs, advice? } } (tell the user the estimate; advice = faster options for a long Whisper run); poll clips_get. Then clips_export, or edit a clip's projectPath with any project tool.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -6476,7 +6601,7 @@ const TOOLS = [
 	{
 		name: "clips_get",
 		description:
-			"One clip set: status (importing, transcribing, finding, building, ready, error), usedModel, sourceProjectPath and clips[] best-first ({ id, title, hook, score 0-100, startMs, endMs, status, projectPath, thumbPath, exportedPath }).",
+			"One clip set: status (importing, transcribing, finding, building, ready, error), stage { message, percent }, transcription { engine, estimate, advice?, etaMs (time left while transcribing) }, usedModel, sourceProjectPath and clips[] best-first ({ id, title, hook, score 0-100, startMs, endMs, status, projectPath, thumbPath, exportedPath }).",
 		inputSchema: {
 			type: "object",
 			properties: { id: { type: "string", description: "Clip set id." } },

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.234.0",
+	"version": "1.241.0",
 	"description": "Model Context Protocol server for PandaStudio. Exposes the desktop video editor's automation surface to Cursor, Continue, Cline, Claude Desktop, and any MCP-compliant client.",
 	"keywords": [
 		"pandastudio",