@writepanda/mcp 1.240.0 → 1.242.0

package/bin/server.mjs

@@ -4499,7 +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: {
@@ -4782,7 +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. 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.",
+			"Remove measured quiet runs inside transcript pauses AND inside measured audio pauses the word timings run across (speech-to-text often stretches a word over the pause after it), never into a word's voiced sound, with a 50ms decay hold. Counts only visible time (nothing under a head trim or existing cut). Returns snapped/skipped/warnings and audioPauses {found, quiet, clearOfWords, cut} (the silencedetect spans at least thresholdMs long, how many the full-band level confirms quiet, how many leave room clear of words, how many were cut). removedCount 0 always comes with a warning saying why: do not run your own silencedetect. 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: {
@@ -6563,7 +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: {
@@ -6601,7 +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": "1.240.0",
+	"version": "1.242.0",
 	"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",