@writepanda/mcp 1.76.0 → 1.81.0

package/bin/server.mjs

@@ -379,7 +379,10 @@ const TOOLS = [
 				id: { type: "string", description: "Export entry id." },
 				accountId: { type: "string" },
 				channelId: { type: "string" },
-				title: { type: "string", description: "Max 100 chars; < and > are stripped (YouTube rejects them)." },
+				title: {
+					type: "string",
+					description: "Max 100 chars; < and > are stripped (YouTube rejects them).",
+				},
 				description: {
 					type: "string",
 					description:
@@ -420,7 +423,10 @@ const TOOLS = [
 			properties: {
 				accountId: { type: "string" },
 				videoId: { type: "string" },
-				title: { type: "string", description: "Max 100 chars; < and > are stripped (YouTube rejects them)." },
+				title: {
+					type: "string",
+					description: "Max 100 chars; < and > are stripped (YouTube rejects them).",
+				},
 				description: {
 					type: "string",
 					description:
@@ -845,10 +851,22 @@ const TOOLS = [
 			properties: {
 				id: { type: "string" },
 				path: { type: "string" },
-				transitionId: { type: "string", description: "Bundled id, e.g. fade-black. From asset_list_transitions." },
-				file: { type: "string", description: "Custom transition video (path / file:// URL). Use instead of transitionId." },
-				atMs: { type: "number", description: "Edited-timeline CUT time to center the transition on." },
-				durationMs: { type: "number", description: "Total length. Default 1000 (native length of the bundled set)." },
+				transitionId: {
+					type: "string",
+					description: "Bundled id, e.g. fade-black. From asset_list_transitions.",
+				},
+				file: {
+					type: "string",
+					description: "Custom transition video (path / file:// URL). Use instead of transitionId.",
+				},
+				atMs: {
+					type: "number",
+					description: "Edited-timeline CUT time to center the transition on.",
+				},
+				durationMs: {
+					type: "number",
+					description: "Total length. Default 1000 (native length of the bundled set).",
+				},
 				expectedRevision: { type: "number" },
 			},
 			required: ["atMs"],
@@ -919,8 +937,16 @@ const TOOLS = [
 					description:
 						"1 (mild) - 6 (extreme). Default 2 (1.5×) — soft, modern feel. Use 3-4 for emphasis beats, 5-6 for dramatic punch-ins.",
 				},
-				focusX: { type: "number", description: "Normalised 0-1 horizontal center. Default 0.5. Ignored when followCursor is true." },
-				focusY: { type: "number", description: "Normalised 0-1 vertical center. Default 0.5. Ignored when followCursor is true." },
+				focusX: {
+					type: "number",
+					description:
+						"Normalised 0-1 horizontal center. Default 0.5. Ignored when followCursor is true.",
+				},
+				focusY: {
+					type: "number",
+					description:
+						"Normalised 0-1 vertical center. Default 0.5. Ignored when followCursor is true.",
+				},
 				followCursor: {
 					type: "boolean",
 					description:
@@ -1053,7 +1079,7 @@ const TOOLS = [
 	{
 		name: "project_add_clip_transform_region",
 		description:
-			"Time-bounded transform on the main video clip between startMs and endMs. TWO families of preset: (1) cam-* render the single clip inside a sub-rect of the canvas, freeing the rest for motion graphics (use ONLY on camera-only / user-uploaded recordings; for screen recordings use cursor-telemetry zooms). (2) layout-* swap the WHOLE podcast composite for the window — this is how you change a PODCAST's layout over time WITHIN ONE recording (e.g. side-by-side intro, then host-only while they talk, then back). Outside the region the clip renders with its natural layout. Renderer interpolates over `transitionMs` (default 320ms) at each boundary so the camera/tiles smoothly resize.",
+			"Time-bounded transform on the main video clip between startMs and endMs. TWO families of preset: (1) cam-* render the single clip inside a sub-rect of the canvas, freeing the rest for motion graphics (use ONLY on camera-only / user-uploaded recordings; for screen recordings use cursor-telemetry zooms). (2) podcast layouts swap the WHOLE podcast composite for the window — change which participant(s) are shown over time WITHIN ONE recording. For multi-party (host + up to 3 guests) use the PARTICIPANT-AWARE presets podcast-solo / podcast-pair / podcast-grid / podcast-screen-share with a `participants` list (e.g. show just guest-2, or Kamal+Vimal side by side). Outside the region the clip renders with its natural layout. Renderer interpolates over `transitionMs` (default 320ms) at each boundary so the tiles smoothly resize.",
 		inputSchema: {
 			type: "object",
 			required: ["startMs", "endMs", "preset"],
@@ -1080,9 +1106,19 @@ const TOOLS = [
 						"layout-podcast",
 						"layout-host-full",
 						"layout-guest-full",
+						"podcast-solo",
+						"podcast-pair",
+						"podcast-grid",
+						"podcast-screen-share",
 					],
 					description:
-						"cam-* (camera reposition into a sub-rect): cam-bottom-half / cam-top-half (9:16); cam-right-portrait / cam-left-portrait (16:9 ~35% card); cam-right-portrait-sm (small ~25% host card — for a full-frame graphic background); cam-bottom-right-quarter / cam-bottom-left-quarter (corner card); cam-right-55/left-55/right-50/left-50 (designed splits). layout-* (PODCAST ONLY — whole-composite swap over time): layout-side-by-side, layout-podcast (two equal tiles), layout-host-full (speaker 1 only), layout-guest-full (speaker 2 only). IMPORTANT: layout-* presets only do anything on a PODCAST clip (kind='podcast' — a clip with BOTH a host and a guest source). On any other clip there is no second speaker, so they are a no-op; use cam-* there. Check clip.kind via project_read before using layout-*.",
+						"cam-* (camera reposition into a sub-rect): cam-bottom-half / cam-top-half (9:16); cam-right-portrait / cam-left-portrait (16:9 ~35% card); cam-right-portrait-sm (small ~25% host card — for a full-frame graphic background); cam-bottom-right-quarter / cam-bottom-left-quarter (corner card); cam-right-55/left-55/right-50/left-50 (designed splits). PODCAST layouts (whole-composite swap over time, PODCAST clips only): the PARTICIPANT-AWARE family is podcast-solo (one person — pass participants=[who]), podcast-pair (two — participants=[left,right]), podcast-grid (everyone, or participants subset), podcast-screen-share (shared screen + participant strip). Legacy 2-person presets layout-side-by-side / layout-podcast / layout-host-full / layout-guest-full still work. IMPORTANT: podcast/layout-* only do anything on a PODCAST clip (kind='podcast'); on other clips use cam-*. Check clip.kind + participants via project_read first.",
+				},
+				participants: {
+					type: "array",
+					items: { type: "string" },
+					description:
+						"Ordered participant speaker ids for the participant-aware podcast-* presets: 'host' | 'guest' | 'guest-2' | 'guest-3'. podcast-solo→[who], podcast-pair→[left,right], podcast-grid/podcast-screen-share→the set to show (omit = everyone). Ignored by cam-* and legacy layout-* presets.",
 				},
 				transitionMs: {
 					type: "number",
@@ -1126,12 +1162,17 @@ const TOOLS = [
 	{
 		name: "project_delete",
 		description:
-			"Permanently delete a project file from disk. There is no trash — this is irreversible. Use project_list to confirm the id before deleting.",
+			"Permanently delete a project file from disk. There is no trash — this is irreversible. By default the original source recording is KEPT (only the project is removed). Pass deleteRecording=true to ALSO permanently delete the project's original recording file(s) from disk. Use project_list to confirm the id before deleting.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string" },
 				path: { type: "string" },
+				deleteRecording: {
+					type: "boolean",
+					description:
+						"Also permanently delete the project's original source recording file(s). Default false. IRREVERSIBLE.",
+				},
 			},
 		},
 		command: "project.delete",
@@ -1692,6 +1733,26 @@ const TOOLS = [
 		},
 		command: "transcript.delete-words",
 	},
+	{
+		name: "transcript_restore_words",
+		description:
+			"Restore previously deleted words — removes the trim region(s) covering them, undoing transcript_delete_words / transcript_remove_fillers / repeat removal. THIS is the editor's 'select struck-through words → Restore' workflow, callable from MCP. Silence-removal trims are left untouched.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				wordIds: {
+					type: "array",
+					items: { type: "string" },
+					description: "Word IDs (from transcript_get) that are currently deleted.",
+				},
+				expectedRevision: { type: "number" },
+			},
+			required: ["wordIds"],
+		},
+		command: "transcript.restore-words",
+	},
 	{
 		name: "transcript_remove_fillers",
 		description:
@@ -1758,7 +1819,7 @@ const TOOLS = [
 				find: {
 					type: "string",
 					description:
-						"Phrase to search for. Case-insensitive; digits and punctuation are ignored on both sides, so \"than60\" and \"than\" both match the word \"than60\". A pure-number/punctuation phrase can't be targeted (it normalizes to empty) — include a letter word.",
+						'Phrase to search for. Case-insensitive; digits and punctuation are ignored on both sides, so "than60" and "than" both match the word "than60". A pure-number/punctuation phrase can\'t be targeted (it normalizes to empty) — include a letter word.',
 				},
 				replace: { type: "string", description: "Replacement text (replaces the whole phrase)" },
 				expectedRevision: { type: "number" },
@@ -1882,7 +1943,8 @@ const TOOLS = [
 				strokeWidth: { type: "number", description: "Text stroke width in px" },
 				fontSize: {
 					type: "string",
-					description: "Font size in REM, e.g. '2.6rem'. Clamped to 1.0–5.0rem (the editor's slider range).",
+					description:
+						"Font size in REM, e.g. '2.6rem'. Clamped to 1.0–5.0rem (the editor's slider range).",
 				},
 				expectedRevision: { type: "number" },
 			},
@@ -1937,6 +1999,45 @@ const TOOLS = [
 		},
 		command: "media.generate-image",
 	},
+	{
+		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).",
+		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.'",
+				},
+				model: {
+					type: "string",
+					description: "'elevenlabs-v3' (default) | '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.",
+				},
+				language: {
+					type: "string",
+					description: "Optional BCP-47 language hint, e.g. 'en-US', 'es-ES'.",
+				},
+				style: {
+					type: "string",
+					description:
+						"Optional tone/emotion hint (e.g. 'warm and upbeat'). Maps to Gemini's style prompt and MiniMax's emotion enum.",
+				},
+				speed: {
+					type: "number",
+					description: "Optional playback speed for MiniMax (0.5-2, 1 = normal).",
+				},
+			},
+			required: ["text"],
+		},
+		command: "media.generate-narration",
+	},
 	// `motion_generate` removed in v1.31.0 — see comment above the
 	// motion-graphics section.
 	{
@@ -2263,20 +2364,38 @@ const TOOLS = [
 	{
 		name: "export_generate_thumbnail",
 		description:
-			"Generate a YouTube thumbnail for an export-library entry via Replicate's gpt-image-2 model. If `prompt` isn't supplied, the local LLM writes one from the transcript (an opinionated YouTube-thumbnail art-director prompt targeting 3:2 aspect, photorealistic or bold-graphic, high contrast, optional short overlay text). Requires the user's Replicate API key to be set in Settings → Integrations (the key is stored encrypted via OS keychain; PandaStudio never sees the user's OpenAI bill). Returns { imagePath, prompt, iterations } on success.",
+			"Generate a YouTube thumbnail for an export-library entry via Replicate's gpt-image-2 model (3:2, photorealistic, high contrast, mobile-readable). Two ways to drive it: (1) pass `subject` (the topic-relevant hero image, 4-10 words) + `hook` (2-4 word overlay) and optionally `reaction`/`layout` — the prompt is assembled deterministically, no transcript needed; or (2) omit them and the local LLM suggests subject/hook/reaction/layout from the transcript. `prompt` overrides everything with a verbatim prompt. `referenceImagePath` is the person's face (passed as input_images). Requires the user's Replicate API key (Settings → Integrations; stored encrypted). Returns { imagePath, prompt, iterations }.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string", description: "Export entry id." },
-				prompt: {
+				subject: {
+					type: "string",
+					description:
+						"The topic-relevant hero image, described concretely (4-10 words), e.g. 'a phone showing a first $79 Stripe sale'. When set, no transcript/LLM is needed.",
+				},
+				hook: {
+					type: "string",
+					description: "2-4 word overlay text (auto-uppercased), e.g. 'FIRST SALE'.",
+				},
+				reaction: {
+					type: "string",
+					description:
+						"Facial reaction: excitement (default) | shock | delight | awe | curiosity | pride | determination | relief.",
+				},
+				layout: {
 					type: "string",
 					description:
-						"Optional explicit prompt. If omitted, the local LLM auto-writes one from the transcript.",
+						"Composition: reaction-split (default) | subject-hero | before-after | big-face | product-only | versus.",
+				},
+				prompt: {
+					type: "string",
+					description: "Optional verbatim prompt. Overrides subject/hook/reaction/layout entirely.",
 				},
 				referenceImagePath: {
 					type: "string",
 					description:
-						"Optional reference image (absolute path or https URL) passed as gpt-image-2 `input_images`.",
+						"Optional reference person image (absolute path or https URL) passed as gpt-image-2 `input_images`.",
 				},
 				quality: {
 					type: "string",

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.76.0",
+	"version": "1.81.0",
 	"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",