@writepanda/mcp 1.285.0 → 1.286.0

package/bin/server.mjs

@@ -1416,7 +1416,7 @@ const TOOLS = [
 	{
 		name: "youtube_list_accounts",
 		description:
-			"Connected YouTube accounts and channels (pick before export_publish_youtube). issues[] = accounts missing a Google permission (code youtube_permission_missing): relay message, follow howToFix (the user reconnects in Settings > YouTube, leaving every box ticked; you can't do the sign-in).",
+			"Connected YouTube accounts and channels (pick before export_publish_youtube). Each channel's userLastPrivacy = what the user last picked in the app's Publish window (use only when they ask for their usual setting). issues[] = accounts missing a Google permission (code youtube_permission_missing): relay message, follow howToFix (the user reconnects in Settings > YouTube, leaving every box ticked; you can't do the sign-in).",
 		inputSchema: { type: "object", properties: {} },
 		command: "youtube.list-accounts",
 	},
@@ -1434,7 +1434,7 @@ const TOOLS = [
 	{
 		name: "export_publish_youtube",
 		description:
-			"Publish an export entry to YouTube. Slow. Check the entry's youtubeVideoId before republishing. channelId optional (uploads go to the account's own channel; omit when channels can't be listed). Error code youtube_permission_missing: relay message, follow howToFix. Error code daily_limit or youtube_quota_exceeded (with resetsAt): tell the user when it resets; never retry before resetsAt. Returns { videoUrl, channelId }.",
+			"Publish an export entry to YouTube. Slow. Check the entry's youtubeVideoId before republishing. channelId optional (uploads go to the account's own channel; omit when channels can't be listed). Error code youtube_permission_missing: relay message, follow howToFix. Error code daily_limit or youtube_quota_exceeded (with resetsAt): tell the user when it resets; never retry before resetsAt. privacyStatus defaults to unlisted: public only when the user asked. A vertical or square export of 3 min or less becomes a YouTube Short (result short: true); pass addShortsTag to add #Shorts. Returns { videoUrl, channelId, short }.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -1447,6 +1447,10 @@ const TOOLS = [
 				categoryId: { type: "string" },
 				privacyStatus: { type: "string", enum: ["public", "unlisted", "private"] },
 				setThumbnail: { type: "boolean" },
+				addShortsTag: {
+					type: "boolean",
+					description: "Add #Shorts to the description when the export is a Short. Default false.",
+				},
 				subtitles: {
 					type: "boolean",
 					description:
@@ -1463,7 +1467,7 @@ const TOOLS = [
 	{
 		name: "social_channels",
 		description:
-			"Social accounts connected in the active workspace (Instagram, TikTok, Facebook, LinkedIn, X) with ids, limits and per-post settings, plus each network's count.",
+			"Social accounts connected in the active workspace (Instagram, TikTok, Facebook, LinkedIn, X) with ids, limits and per-post settings, plus each network's count. X accounts carry xPremium (25,000-char limit when true, else 280). YouTube and YouTube Shorts aren't here: use youtube_list_accounts + export_publish_youtube.",
 		inputSchema: { type: "object", properties: {} },
 		command: "social.channels",
 	},
@@ -1490,6 +1494,20 @@ const TOOLS = [
 		},
 		command: "social.connect-status",
 	},
+	{
+		name: "social_set_x_premium",
+		description:
+			"Mark a connected X account as X Premium (posts up to 25,000 chars) or standard (280). Only when the user says the account has Premium: X rejects long posts from other accounts. Returns the updated channel.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				channelId: { type: "string", description: "The X account's id from social_channels." },
+				premium: { type: "boolean" },
+			},
+			required: ["channelId", "premium"],
+		},
+		command: "social.set-x-premium",
+	},
 	{
 		name: "social_disconnect",
 		description:
@@ -1803,7 +1821,7 @@ const TOOLS = [
 	{
 		name: "project_edit_summary",
 		description:
-			"What the edit contains now: lengths (editedDurationMs, sourceDurationMs, removedMs cut or sped up), counts (cuts, zooms, motionElements, graphics, transitions, brollBeats, mediaOverlays, captions { on, template }, audio { narration, music, soundEffects }, voiceEnhancedClips vs loudnessNormalizedClips, ...) and highlights[] (plain lines). Each recording counts once, so splits change neither source length nor cuts. Cheap; use it to report an editing pass instead of project_read.",
+			"What the edit contains now: lengths (editedDurationMs, sourceDurationMs, removedMs cut or sped up), counts (cuts, zooms, motionElements, graphics, transitions, brollBeats, mediaOverlays, captions { on, template }, audio { narration, music, soundEffects }, audioTracks[{id, kind voice|music|sfx, basis}] (the same kind the export mix and audio_preview_mix use), voiceEnhancedClips vs loudnessNormalizedClips, ...) and highlights[] (plain lines). Each recording counts once, so splits change neither source length nor cuts. Cheap; use it to report an editing pass instead of project_read.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -2092,14 +2110,17 @@ const TOOLS = [
 		],
 		[
 			"stop-recording",
-			"Stop recording and finalize podcast media. Check podcast_status for finalization and guest uploads.",
+			"Stop recording. The take goes into the episode project by itself (desktop track once finalized, each guest's as their upload lands, even after the studio is left); returns savedTo and takeId. podcast_status shows finalization and guest uploads.",
 		],
 		[
 			"start-screen-share",
 			"Open the desktop picker to share the host screen with everyone in the call (any time, backstage included; it's recorded while recording). One sharer at a time. Waits for the host to choose; fails if the picker is cancelled or capture fails.",
 		],
 		["stop-screen-share", "Stop and finalize the host's podcast screen share."],
-		["open-in-editor", "Open the finished podcast in the editor when guest media has arrived."],
+		[
+			"open-in-editor",
+			"Open the finished podcast's episode in the editor when guest media has arrived (the take is already in the episode project).",
+		],
 	].map(([noun, description]) => ({
 		name: `podcast_${noun.replaceAll("-", "_")}`,
 		description,
@@ -2623,6 +2644,34 @@ const TOOLS = [
 		},
 		command: "podcast.list-upcoming",
 	},
+	{
+		name: "podcast_found_recordings",
+		description:
+			"Recordings that belong to a podcast episode but never reached it (older takes left by Back before Open in editor): guest tracks matched by session code plus the nearest desktop camera recording, each with a key. No path / id / sessionCode = every episode.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				path: { type: "string" },
+				id: { type: "string" },
+				sessionCode: { type: "string" },
+			},
+		},
+		command: "podcast.found-recordings",
+	},
+	{
+		name: "podcast_add_found_recordings",
+		description:
+			"Add a podcast_found_recordings take (path + key) to its episode: it becomes the episode's podcast clip, files move into the project folder. Matched by time, so ask the user first; check the sync afterwards (project_auto_sync_podcast).",
+		inputSchema: {
+			type: "object",
+			properties: {
+				path: { type: "string", description: "The episode's project file." },
+				key: { type: "string", description: "The found take's key." },
+			},
+			required: ["path", "key"],
+		},
+		command: "podcast.add-found-recordings",
+	},
 	{
 		name: "podcast_get_episode",
 		description:
@@ -3698,6 +3747,39 @@ const TOOLS = [
 		},
 		command: "project.remove-mute-region",
 	},
+	{
+		name: "project_add_original_voice",
+		description:
+			'Turn Enhance voice off for an edited-time span ("Original voice"): the main track plays its original audio there, enhanced elsewhere. Only enhanced clips change. Returns regionId (editor.enhanceOffRegions).',
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				atMs: { type: "number" },
+				durationMs: { type: "number" },
+				expectedRevision: { type: "number" },
+			},
+			required: ["atMs", "durationMs"],
+		},
+		command: "project.add-original-voice",
+	},
+	{
+		name: "project_remove_original_voice",
+		description:
+			"Remove an Original voice region so Enhance voice applies there again (ids: editor.enhanceOffRegions[].id).",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				regionId: { type: "string" },
+				expectedRevision: { type: "number" },
+			},
+			required: ["regionId"],
+		},
+		command: "project.remove-original-voice",
+	},
 	{
 		name: "project_add_bleep",
 		description:
@@ -4143,7 +4225,7 @@ const TOOLS = [
 	{
 		name: "project_review_moment",
 		description:
-			"LOOK AND LISTEN to a moment of the edit as the viewer gets it: a contact sheet of the frames around atMs (edited ms) or wordId, the EXPORTED mix of that stretch transcribed and compared with what the edit should play (extra = leftover of a cut word or line, missing = clipped word), whether each cut lands in quiet or speech, nearby edits and stutters, plus a plain verdict. Use after every cut / pause / insert fix BEFORE telling the user it's fixed, and when the user names a moment ('at 16 seconds'). ASYNC: { jobId }; job_wait.",
+			"LOOK AND LISTEN to a moment of the edit as the viewer gets it: a contact sheet of the frames around atMs (edited ms) or wordId, the EXPORTED mix of that stretch transcribed and compared with what the edit should play (the clips' words AND narration / voiceover track words; extra = leftover of a cut word or line, missing = clipped word), whether each cut that interrupts the main track's sound lands in quiet or speech (judged on the main track alone, so narration or music playing across it doesn't count; a join of silent clips such as images is never a cut in the speech), nearby edits and stutters, plus a plain verdict. Use after every cut / pause / insert fix BEFORE telling the user it's fixed, and when the user names a moment ('at 16 seconds'). ASYNC: { jobId }; job_wait.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -4198,7 +4280,7 @@ const TOOLS = [
 	{
 		name: "audio_preview_mix",
 		description:
-			"Render final export audio only to temporary .m4a plus description and voice/music/SFX stem RMS/masking flags. Edited-time range fromMs..toMs inside 0..editedDurationMs (project_read), default whole project, max 10 minutes. Async: job.wait.",
+			"Render final export audio only to temporary .m4a plus description and voice/music/SFX stem RMS/masking flags; tracks[{id, kind, basis}] = which stem each audio track went to (same as project_edit_summary audioTracks). Edited-time range fromMs..toMs inside 0..editedDurationMs (project_read), default whole project, max 10 minutes. Async: job.wait.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -5126,7 +5208,7 @@ const TOOLS = [
 	{
 		name: "project_auto_reframe",
 		description:
-			"Active-speaker auto-reframe: frames whoever talks, shot by shot (face tracking, no speaker IDs needed). Landscape footage on a vertical / square canvas gets a crop that follows the speaker; footage that already has the canvas shape (a 16:9 wide of two hosts in a 16:9 video) gets a punch-in on the speaker inside the frame, which is also how to get face-tracked zooms on the main video. zoom = fixed punch-in. split=true stacks two people (podcast / interview wide) one above the other on a 9:16 canvas. Slow (minutes). clear reverts. Concurrent saves are merged and retried (three attempts). Exhausted conflicts report code revision_conflict / status 409 (job.errorDetails for async jobs).",
+			"Active-speaker auto-reframe: frames whoever talks, shot by shot (face tracking, no speaker IDs needed). Landscape footage on a vertical / square canvas gets a crop that follows the speaker; footage that already has the canvas shape (a 16:9 wide of two hosts in a 16:9 video) gets a punch-in on the speaker inside the frame, which is also how to get face-tracked zooms on the main video. zoom = fixed punch-in. split=true stacks two people (podcast / interview wide) one above the other on a 9:16 canvas (not on a clip with its own camera video). In the vertical-stack camera layout the crop is computed for the main video's top half. Slow (minutes). clear reverts. 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: {
@@ -5396,6 +5478,12 @@ const TOOLS = [
 					type: "boolean",
 					description: "Repeat the file to endMs (a ~30 s Lyria clip under a longer video).",
 				},
+				audioKind: {
+					type: "string",
+					enum: ["voice", "music", "sfx"],
+					description:
+						"What the track is (its mix bus, preview-mix stem and edit-summary count). Omit for bundled library files and generated narration / music / SFX (known); pass it for other files, e.g. music for an uploaded song. Change later: project_update_region regionType=audio-overlay patch {audioKind}.",
+				},
 				transcribe: {
 					type: "boolean",
 					description:
@@ -5807,7 +5895,7 @@ const TOOLS = [
 	{
 		name: "project_set_clip_webcam",
 		description:
-			"Attach, replace or remove a clip's camera video (Screen + camera), in sync from its first frame. Not for podcast composites.",
+			"Attach, replace or remove a clip's camera video, in sync from its first frame: a presenter camera over a screen recording, or a REACTION video (main clip = the video reacted to, camera = the reactor; audio=mix plays both). The main video keeps its kind. In 9:16 use set-webcam-layout vertical-stack (main on top, camera below), then auto-reframe frames the main video inside its half. Not for podcast composites.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -5818,8 +5906,9 @@ const TOOLS = [
 				webcam: { type: "string", description: "Absolute path, or '' to remove." },
 				audio: {
 					type: "string",
-					enum: ["auto", "camera", "keep"],
-					description: "auto: camera audio if the main video is silent (re-transcribe after).",
+					enum: ["auto", "mix", "camera", "keep"],
+					description:
+						"auto: camera audio if the main video is silent; mix: both sounds (reaction video); camera: camera's sound only; keep: unchanged. Re-transcribe after a change.",
 				},
 				expectedRevision: { type: "number" },
 			},
@@ -5895,13 +5984,14 @@ const TOOLS = [
 	{
 		name: "system_set_transcription_preferences",
 		description:
-			"Save workspace transcription language, provider and/or on-device engine atomically (pass any). engine: auto (Parakeet for English/auto-detect, Whisper LOCKED to any other named language), parakeet (keeps fillers, detects the language, can't be locked), whisper (locks to the language). Confirm consent to upload audio for cloud. Cloud and apple cannot use auto.",
+			"Save workspace transcription language, provider and/or on-device engine atomically (pass any). engine: auto (Parakeet for English/auto-detect, Whisper LOCKED to any other named language), parakeet (keeps fillers, detects the language, can't be locked), whisper (locks to the language). Confirm consent to upload audio for cloud. Cloud and apple cannot use auto. fallback: when the cloud provider / Apple speech still fails after retries, local (default: this computer transcribes only the failed parts, reported) or none (fail, never transcribe on this computer: use when the user wants the cloud provider only).",
 		inputSchema: {
 			type: "object",
 			properties: {
 				language: { type: "string" },
 				provider: { type: "string", enum: ["local", "apple", "deepgram", "elevenlabs"] },
 				engine: { type: "string", enum: ["auto", "parakeet", "whisper"] },
+				fallback: { type: "string", enum: ["local", "none"] },
 			},
 		},
 		command: "system.set-transcription-preferences",
@@ -5910,13 +6000,19 @@ const TOOLS = [
 	{
 		name: "transcript_transcribe",
 		description:
-			"Transcribe clips with word timestamps (default: only clips without words). Asking again while the same transcription is still running returns that job (alreadyRunning: true), never a second run: keep waiting on it. 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[]). Clips with no audio track are skipped (result noAudioClips), not failed. Result warnings[] flags words that can't cover the audio (sound with no words, too few words: music, noise, another language) and a re-run with the same words as before: surface it, and get a second reading with engine whisper or a cloud provider (Parakeet isn't locked by language english, so retrying the same way repeats it). 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). Asking again while the same transcription is still running returns that job (alreadyRunning: true), never a second run: keep waiting on it. 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[]). Clips with no audio track are skipped (result noAudioClips), not failed. Result warnings[] flags words that can't cover the audio (sound with no words, too few words: music, noise, another language) and a re-run with the same words as before: surface it, and get a second reading with engine whisper or a cloud provider (Parakeet isn't locked by language english, so retrying the same way repeats it). 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). CLOUD (deepgram / elevenlabs): long audio goes up in ~5 min parts, each retried on network drops, timeouts, 5xx and 429. fallback: none = parts that still fail FAIL the job (job.errorDetails.code cloud_failed, failedRanges; nothing runs on this computer; parts that worked are kept for a re-run), local = this computer transcribes only the failed parts. Default none when you pass a cloud/apple provider, else the workspace setting. The job says which engine really ran: progress.data.engine / fallbackFrom while running, result clips[].engine / engines / fallback. job_cancel stops every upload, retry and on-device helper of the job.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string" },
 				path: { type: "string" },
 				clipId: { type: "string", description: "Only this clip, even if already transcribed." },
+				fallback: {
+					type: "string",
+					enum: ["local", "none"],
+					description:
+						"When the cloud provider / Apple speech still fails after retries: none = fail the job (never transcribe on this computer), local = this computer transcribes only the failed parts. Default: none when provider is passed, else the workspace setting.",
+				},
 				language: {
 					type: "string",
 					description:
@@ -8676,13 +8772,18 @@ const TOOLS = [
 	{
 		name: "llm_generate_title",
 		description:
-			"Generate a YouTube title from what the EDITED video says (cut words left out) with the local LLM. Returns { title }: one complete phrase within maxChars, never cut mid-word.",
+			"Generate a YouTube title from what the EDITED video says (cut words left out) with the local LLM. Names the people speaking when the project knows them (podcast participants); instruction steers it. Returns { title }: one complete phrase within maxChars, never cut mid-word.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string" },
 				path: { type: "string" },
 				maxChars: { type: "number" },
+				instruction: {
+					type: "string",
+					description:
+						"Optional steer from the user (what to focus on), max 500 chars. Speaker names the project knows are used automatically.",
+				},
 			},
 		},
 		command: "llm.generate-title",
@@ -8690,13 +8791,18 @@ const TOOLS = [
 	{
 		name: "llm_generate_description",
 		description:
-			"Generate a 3-5 sentence YouTube description from what the EDITED video says, with the local LLM. Returns { description, warnings? }: a warning means it repeated the transcript, so write the description yourself.",
+			"Generate a 3-5 sentence YouTube description from what the EDITED video says, with the local LLM. Names the people speaking when the project knows them (podcast participants); instruction steers what it focuses on. Returns { description, warnings? }: a warning means it repeated the transcript, so write the description yourself.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string" },
 				path: { type: "string" },
 				maxChars: { type: "number" },
+				instruction: {
+					type: "string",
+					description:
+						"Optional steer from the user (what to focus on), max 500 chars. Speaker names the project knows are used automatically.",
+				},
 			},
 		},
 		command: "llm.generate-description",
@@ -8725,6 +8831,11 @@ const TOOLS = [
 				id: { type: "string" },
 				path: { type: "string" },
 				maxChapters: { type: "number" },
+				instruction: {
+					type: "string",
+					description:
+						"Optional steer from the user (what to focus on), max 500 chars. Speaker names the project knows are used automatically.",
+				},
 			},
 		},
 		command: "llm.generate-timestamps",
@@ -8939,7 +9050,7 @@ const TOOLS = [
 	{
 		name: "clips_export",
 		description:
-			"Export a clip set's finished clips to MP4 (all, or clipIds) into Movies/PandaStudio Clips/<video>/, at standard quality (1080p; 720p on a free trial). Async: returns { jobId, outputDir }; job_wait for { exported[] }.",
+			"Export a clip set's finished clips to MP4 (all, or clipIds) into Movies/PandaStudio Clips/<video>/, at standard quality (1080p; 720p on a free trial). Async: returns { jobId, outputDir }; job_wait for { exported[{ clipId, outputPath, exportId }] }. To publish a clip, export it here, then pass exportId to export_publish_youtube (9:16 = Short) or export_publish_social.",
 		inputSchema: {
 			type: "object",
 			properties: {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.285.0",
+	"version": "1.286.0",
 	"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",