@writepanda/mcp 1.98.0 → 1.100.0

package/bin/server.mjs

@@ -229,6 +229,40 @@ const TOOLS = [
 		inputSchema: { type: "object", properties: {} },
 		command: "system.isWhisperModelDownloaded",
 	},
+	{
+		name: "system_get_narration_engine",
+		description:
+			"Read the active narration (TTS) engine for the current workspace. Returns { engine } — 'local-kokoro' (default; Kokoro-82M on-device, English, no key/cloud) or 'replicate' (cloud TTS, needs the user's Replicate key). Use before recommending a switch or when disclosing which engine narration uses.",
+		inputSchema: { type: "object", properties: {} },
+		command: "system.getNarrationEngine",
+	},
+	{
+		name: "system_set_narration_engine",
+		description:
+			"Set the DEFAULT narration engine for the current workspace. Pass `engine` = 'local-kokoro' (on-device English, no key) or 'replicate' (cloud, needs the user's Replicate key). Individual media_generate_narration calls can still override this via their `model` arg. Switching to local triggers a one-time ~330 MB model download on first use — call system_is_kokoro_model_downloaded to check.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				engine: { type: "string", description: "'local-kokoro' | 'replicate'." },
+			},
+			required: ["engine"],
+		},
+		command: "system.setNarrationEngine",
+	},
+	{
+		name: "system_is_kokoro_model_downloaded",
+		description:
+			"Check whether the local Kokoro-82M TTS model (English on-device narration) has been downloaded. Returns { downloaded: boolean }. The model (~330 MB) auto-downloads on first local narration; calling this first lets you surface the download instead of a long first-run wait.",
+		inputSchema: { type: "object", properties: {} },
+		command: "system.isKokoroModelDownloaded",
+	},
+	{
+		name: "system_download_kokoro_model",
+		description:
+			"Pre-download the local Kokoro-82M TTS model + voices (~330 MB) now so the first narration doesn't pay the wait. Idempotent; returns { downloaded: true }. Optional — narration also auto-downloads on first use.",
+		inputSchema: { type: "object", properties: {} },
+		command: "system.downloadKokoroModel",
+	},
 
 	// ── workspaces (v1.19) ──────────────────────────────────────────
 	// Multi-workspace isolation: each workspace has its own projects,
@@ -854,7 +888,7 @@ const TOOLS = [
 	{
 		name: "project_add_motion_graphic",
 		description:
-			"Drop a motion-graphic MP4/WebM onto the timeline. PREFERRED: pass `fromJob` (the jobId from motion_render_html / motion_generate / motion_concat) — the server resolves the outputPath internally, so you never handle the path string at all. Fallback: pass `file` (an absolute path) only for external/uploaded files. Also accepts `file: 'bundled:transition/<id>'` (ids from asset_list_transitions) — the house-opener pattern: a bundled transition placed as an overlay auto-stamps transitionId so it COVER-fits any canvas (a 16:9 sweep fills a 9:16 frame) and carries its SFX via soundUrl. **Defaults to a `bundled:sound/mouse-click` SFX as the graphic appears** — pass `soundUrl: null` to make it silent, or any other `bundled:sound/<id>` to swap the sound.",
+			"Drop a motion-graphic MP4/WebM onto the timeline. PREFERRED: pass `fromJob` (the jobId from motion_render_html / motion_generate / motion_concat) — the server resolves the outputPath internally, so you never handle the path string at all. Fallback: pass `file` (an absolute path) only for external/uploaded files. Also accepts `file: 'bundled:transition/<id>'` (ids from asset_list_transitions) — the house-opener pattern: a bundled transition placed as an overlay auto-stamps transitionId so it COVER-fits any canvas (a 16:9 sweep fills a 9:16 frame) and carries its SFX via soundUrl. Image files (png/jpg/webp/gif/avif) are detected and render as image overlays; pair `file: <image>` with `layer: 'background'` for a timed BACKGROUND-IMAGE underlay (cover-fills the canvas behind the camera — with a background-effect remove region over the same span, the image replaces the speaker's real background for just that duration). **Defaults to a `bundled:sound/mouse-click` SFX as the graphic appears** — pass `soundUrl: null` to make it silent, or any other `bundled:sound/<id>` to swap the sound.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -1223,6 +1257,48 @@ const TOOLS = [
 		},
 		command: "project.add-spotlight",
 	},
+	{
+		name: "project_add_background_effect",
+		description:
+			"Add a BACKGROUND EFFECT region on the camera video: AI person segmentation runs between startMs and endMs and either BLURS the background behind the speaker (mode 'blur' — video-call style) or REMOVES it so the project wallpaper shows through behind them (mode 'remove'). Behaves exactly like a zoom region on the timeline: draggable, trimmable, source-anchored (pass anchorSourceMs when placing from a transcript word so it survives later trims). Camera / talking-head footage only. Preview and export render it identically. Find placed regions via project_read under editor.backgroundEffectRegions[]; retime/restyle with project_update_region (regionType 'background-effect'), delete with project_remove_region.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				atMs: { type: "number", description: "Edited-time start (ms). Or pass startMs." },
+				startMs: { type: "number", description: "Alias for atMs (edited-time start, ms)." },
+				endMs: { type: "number", description: "Edited-time end (ms). Alternative to durationMs." },
+				durationMs: {
+					type: "number",
+					description: "Duration in ms. Default 5000 when neither durationMs nor endMs given.",
+				},
+				mode: {
+					type: "string",
+					enum: ["blur", "remove"],
+					description:
+						'"blur" = person sharp, background gaussian-blurred. "remove" = background cut away; the project wallpaper/background shows through behind the person.',
+				},
+				strength: {
+					type: "number",
+					description:
+						"Blur strength in px sigma at 1080p (mode='blur' only; ignored for 'remove'). 1-120, default 18.",
+				},
+				anchorSourceMs: {
+					type: "number",
+					description:
+						"Anchor to a SOURCE-time moment (transcript word startMs). The region then re-anchors automatically on subsequent trim/speed edits.",
+				},
+				anchorSourceEndMs: {
+					type: "number",
+					description: "Optional anchor end (source ms) for ranged anchoring.",
+				},
+				expectedRevision: { type: "number" },
+			},
+			required: ["mode"],
+		},
+		command: "project.add-background-effect",
+	},
 	{
 		name: "project_update_spotlight",
 		description:
@@ -1624,7 +1700,7 @@ const TOOLS = [
 	{
 		name: "project_remove_region",
 		description:
-			"Delete an existing region (zoom, trim, speed, annotation, fx, overlay, clip-transform, audio-overlay) by its id. Use project_read to find region ids — visual regions live under editor.*Regions; audio overlays live under audioOverlays[]. If the region belongs to a link group (designed-segment pair: a clip-transform parking the camera into one half + the panel motion-graphic on the other half), deleting either half also removes its peer so the pair stays coherent. Returns the updated project.",
+			"Delete an existing region (zoom, trim, speed, annotation, fx, overlay, clip-transform, background-effect, audio-overlay) by its id. Use project_read to find region ids — visual regions live under editor.*Regions; audio overlays live under audioOverlays[]. If the region belongs to a link group (designed-segment pair: a clip-transform parking the camera into one half + the panel motion-graphic on the other half), deleting either half also removes its peer so the pair stays coherent. Returns the updated project.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -1633,7 +1709,7 @@ const TOOLS = [
 				regionType: {
 					type: "string",
 					description:
-						"zoom | trim | speed | annotation | fx | overlay | clip-transform | audio-overlay",
+						"zoom | trim | speed | annotation | fx | overlay | clip-transform | background-effect | audio-overlay",
 				},
 				regionId: { type: "string", description: "id of the region to delete" },
 				expectedRevision: { type: "number" },
@@ -1664,7 +1740,7 @@ const TOOLS = [
 	{
 		name: "project_update_region",
 		description:
-			"Patch fields on an existing region without replacing it. Accepts any subset of the region's own properties — only the supplied fields are changed, everything else is preserved. Overlay geometry (x/y/width/height) is PERCENT of canvas 0-100; values <= 1 are treated as 0-1 fractions and scaled. Works on zoom (depth, focusX, focusY, focusMode: 'static'|'auto' — set 'auto' for the cursor-follow camera), trim (startMs, endMs), speed (startMs, endMs, speed), fx (atMs, durationMs, opacity, blendMode, speed), annotation, overlay, clip-transform (startMs, endMs, preset, transitionMs), and audio-overlay (startMs, endMs, sourceStartMs, volume — lets you drag-trim an audio overlay without removing it). Time shifts on a link-grouped region (e.g. one half of a designed segment) propagate to every peer in the group so the pair stays coherent.",
+			"Patch fields on an existing region without replacing it. Accepts any subset of the region's own properties — only the supplied fields are changed, everything else is preserved. Overlay geometry (x/y/width/height) is PERCENT of canvas 0-100; values <= 1 are treated as 0-1 fractions and scaled. Works on zoom (depth, focusX, focusY, focusMode: 'static'|'auto' — set 'auto' for the cursor-follow camera), trim (startMs, endMs), speed (startMs, endMs, speed), fx (atMs, durationMs, opacity, blendMode, speed), annotation, overlay, clip-transform (startMs, endMs, preset, transitionMs), background-effect (startMs, endMs, mode: 'blur'|'remove', strength), and audio-overlay (startMs, endMs, sourceStartMs, volume — lets you drag-trim an audio overlay without removing it). Time shifts on a link-grouped region (e.g. one half of a designed segment) propagate to every peer in the group so the pair stays coherent.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -1673,7 +1749,7 @@ const TOOLS = [
 				regionType: {
 					type: "string",
 					description:
-						"zoom | trim | speed | annotation | fx | overlay | clip-transform | audio-overlay",
+						"zoom | trim | speed | annotation | fx | overlay | clip-transform | background-effect | audio-overlay",
 				},
 				regionId: { type: "string", description: "id of the region to update" },
 				patch: {
@@ -2390,27 +2466,29 @@ const TOOLS = [
 	{
 		name: "media_generate_narration",
 		description:
-			"Generate a voiceover / narration audio clip via Replicate TTS, for promo videos and explainers. Project-agnostic; writes audio to <userData>/narration/ and returns { audioPath, durationMs, model, voice, predictionId }. Canonical workflow: media_generate_narration → project_add_audio (place at startMs, pass the returned durationMs so the overlay is sized to the speech). Models: 'elevenlabs-v3' (default, most expressive — embed inline delivery tags in the text like [excited] or [whispers]) | 'gemini-flash-tts' (30 voices, strong multilingual; 'style' sets tone) | 'minimax-turbo' (fast; 'style' maps to an emotion, 'speed' 0.5-2). Voices — gemini e.g. Kore/Puck/Charon; minimax e.g. Friendly_Person/Wise_Woman/Casual_Guy; omit voice for the model default. Requires the user's Replicate API key (Settings → Integrations).",
+			"Generate a voiceover / narration audio clip. Project-agnostic; writes audio to <userData>/narration/ and returns { audioPath, durationMs, model, voice, predictionId }. Canonical workflow: media_generate_narration → project_add_audio (place at startMs, pass the returned durationMs so the overlay is sized to the speech). DEFAULT is the LOCAL, on-device Kokoro engine (English) — no API key, no cloud, runs on the user's machine; the model auto-downloads (~330 MB) on first use. Kokoro voices (pass in `voice`): af_heart (default), af_bella, am_michael, bf_emma, bm_george, etc. (a*=American, b*=British; f=female, m=male). To use CLOUD TTS instead (more voices/languages, needs the user's Replicate key), set `model` to a Replicate model: 'elevenlabs-v3' (most expressive — embed inline tags like [excited]/[whispers]) | 'gemini-flash-tts' (30 voices, multilingual, 'style' sets tone) | 'minimax-turbo' (fast, 'style' maps to an emotion, 'speed' 0.5-2). Cloud voices — gemini Kore/Puck/Charon; minimax Friendly_Person/Wise_Woman. Omit `model` to honour the workspace default (system_get_narration_engine).",
 		inputSchema: {
 			type: "object",
 			properties: {
 				text: {
 					type: "string",
 					description:
-						"The script to speak. For elevenlabs-v3 you can embed delivery tags inline, e.g. 'Welcome [excited] to the future of editing.'",
+						"The script to speak. For the cloud elevenlabs-v3 model you can embed delivery tags inline, e.g. 'Welcome [excited] to the future of editing.' (Local Kokoro ignores such tags.)",
 				},
 				model: {
 					type: "string",
-					description: "'elevenlabs-v3' (default) | 'gemini-flash-tts' | 'minimax-turbo'.",
+					description:
+						"Engine/model selector. Omit for the workspace default (local Kokoro). 'kokoro-local' forces on-device English. Cloud (needs Replicate key): 'elevenlabs-v3' | 'gemini-flash-tts' | 'minimax-turbo'.",
 				},
 				voice: {
 					type: "string",
 					description:
-						"Voice id/name for the chosen model (gemini: Kore/Puck/...; minimax: Friendly_Person/...). Omit for the model default.",
+						"Voice for the chosen engine. Local Kokoro: af_heart/af_bella/am_michael/bf_emma/bm_george/... Cloud: gemini Kore/Puck/...; minimax Friendly_Person/... Omit for the engine's default voice.",
 				},
 				language: {
 					type: "string",
-					description: "Optional BCP-47 language hint, e.g. 'en-US', 'es-ES'.",
+					description:
+						"Optional BCP-47 language hint for the cloud models, e.g. 'en-US', 'es-ES'. Local Kokoro is English-only and ignores this.",
 				},
 				style: {
 					type: "string",
@@ -3012,7 +3090,7 @@ const TOOLS = [
 const server = new Server(
 	{
 		name: "pandastudio",
-		version: "1.71.0",
+		version: "1.73.0",
 	},
 	{
 		capabilities: {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.98.0",
+	"version": "1.100.0",
 	"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",