@writepanda/mcp 1.259.0 → 1.262.0

package/bin/server.mjs

@@ -35,7 +35,13 @@ import path from "node:path";
 import process from "node:process";
 import { Server } from "@modelcontextprotocol/sdk/server/index.js";
 import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
-import { CallToolRequestSchema, ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js";
+import {
+	CallToolRequestSchema,
+	ListResourcesRequestSchema,
+	ListResourceTemplatesRequestSchema,
+	ListToolsRequestSchema,
+	ReadResourceRequestSchema,
+} from "@modelcontextprotocol/sdk/types.js";
 import {
 	appLaunchEnv,
 	callWithReconnect,
@@ -47,6 +53,14 @@ import {
 	IN_APP_UNAVAILABLE,
 	isInAppCaller,
 } from "./appConnection.mjs";
+import {
+	APP_TOOLS,
+	clientSupportsUi,
+	listUiResources,
+	MediaRegistry,
+	readUiResource,
+	toolUiMeta,
+} from "./apps/index.mjs";
 import { portableSchema } from "./portableSchema.mjs";
 
 // ── Credentials + transport ───────────────────────────────────────────
@@ -1705,6 +1719,19 @@ const TOOLS = [
 		},
 		command: "project.show",
 	},
+	{
+		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 }, ...) and highlights[] (plain lines). Cheap; use it to report an editing pass instead of project_read.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+			},
+		},
+		command: "project.edit-summary",
+	},
 	{
 		name: "matte_status",
 		description:
@@ -3045,7 +3072,7 @@ const TOOLS = [
 					type: "boolean",
 					description: "Follow the recorded cursor (screen recordings).",
 				},
-				soundUrl: { type: "string", description: 'Default bundled:sound/swoosh-fast; "none".' },
+				soundUrl: { type: "string", description: 'Default bundled:sound/motion-swoosh-up; "none".' },
 				soundVolume: { type: "number", description: "0-1." },
 				anchorSourceMs: ANCHOR_PROP,
 				anchorSourceEndMs: ANCHOR_END_PROP,
@@ -6315,7 +6342,7 @@ const TOOLS = [
 	{
 		name: "project_compose_soundtrack",
 		description:
-			"Score this edit: music + sound design composed FROM THE PROJECT'S OWN TIMELINE (zooms, graphics, pops, cuts, speech, stressed words) and placed under the voice with ducking; takes over zoom/overlay default sounds; replaces the previous score on re-run. Call LAST, after cuts/zooms/graphics. dryRun returns the score for media_compose_soundtrack. Returns { audioPath, scorePath, events, heroMoments, acts, soundsTakenOver, replaced }.",
+			"Score this edit: music composed FROM THE PROJECT'S OWN TIMELINE (zooms, graphics, pops, cuts, speech, stressed words) placed under the voice with ducking, plus sound effects as recorded library cues (group score-sfx, sparse under a voice); takes over zoom/overlay/graphic sounds; replaces the previous score and its cues on re-run. Call LAST, after cuts/zooms/graphics, and again after changing them (project_read composedScore.stale / doubledSounds). dryRun returns the score and sfxCues. Returns { audioPath, scorePath, events, heroMoments, acts, soundsTakenOver, sfxCues, paperSfx, replaced }.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -6327,7 +6354,8 @@ const TOOLS = [
 				sfx: { type: "boolean", description: "Default true." },
 				useSpeech: {
 					type: "boolean",
-					description: "Hero hits on the most stressed words. Default true.",
+					description:
+						"Hits on the most stressed words even with nothing on screen. Default true for promo only.",
 				},
 				cues: {
 					type: "array",
@@ -7173,13 +7201,17 @@ const TOOLS = [
 	{
 		name: "recipe_pick",
 		description:
-			"No style named ('edit my video', 'make this a Short')? The recipes that fit this project's format and footage, each with pickWhen, plus defaultId. Choose by what's said, then recipe_apply_style + recipe_render.",
+			"No style named ('edit my video', 'make this a Short')? The recipes that fit this project's format and footage, each with pickWhen, plus defaultId. Choose by what's said (defaultId when nothing clearly fits, never a plain clean-up), then recipe_apply_style + recipe_render. A Short cut from a landscape long-form video (a project_fork_from_shot fork, a landscape source, or format short on a 16:9 project) comes back fromLongForm with its own default. On a 16:9 project the user wants as a Short, pass format short.",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string", description: "Project id (default: the open project)." },
 				path: { type: "string" },
-				format: { type: "string", enum: ["short", "long"] },
+				format: {
+					type: "string",
+					enum: ["short", "long"],
+					description: "Default from the project's aspect ratio (9:16 = short).",
+				},
 				footage: {
 					type: "string",
 					enum: ["camera", "screen", "podcast", "audio", "none"],
@@ -8001,6 +8033,24 @@ const TOOLS = [
 		},
 		command: "clips.delete",
 	},
+	{
+		name: "clips_remove_clip",
+		description:
+			"Remove ONE clip from a clip set (the Clips screen's Remove on a clip card). Its stills go and its moment is remembered so clips.retry won't make it again; its project moves to the Trash (recoverable) unless keepProject is true. Exported files stay in My Exports. A clip still being made is dropped by the running job. Returns { clipSetId, removed, projectDeleted, clipsLeft }.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string", description: "Clip set id." },
+				clipId: { type: "string", description: "The clip's id (from clips_get)." },
+				keepProject: {
+					type: "boolean",
+					description: "Keep the clip's project in Projects instead of the Trash. Default false.",
+				},
+			},
+			required: ["id", "clipId"],
+		},
+		command: "clips.remove-clip",
+	},
 	{
 		name: "ideas_research",
 		description:
@@ -8654,8 +8704,9 @@ Before an editing pass, read the skill's "Which tool for which moment" section (
 3. Fix misheard names / brands with transcript_find_replace (transcript_insert_words for dropped words).
 4. transcript_find_issues, then transcript_delete_words on each duplicate take / false start (keep the LAST take; keep severity "low" ones).
 5. transcript_remove_silences.
+5b. A Short from a landscape video (a long-form recording, the project isn't 9:16): cut the Short out now. export_generate_shots { projectId }, then project_fork_from_shot on the best shot (offer the top shots when the user is there) and run steps 6-9 on the fork. Only a clip already under about 90 s goes vertical in place (project_set_aspect_ratio 9:16).
 6. audio_enhance (Enhance voice) on clips not yet enhanced with its defaults (Light, plus Studio mic sound on Apple silicon; don't pass strength or restore unless the user asked); keepBackground true for vlogs / ambience; no echo (Enhance voice already removes room echo with the Studio AI model; only use echo when the result reports the model did not run (fallback)).
-7. The style, after steps 2-5 (they shift the timeline). The user's named style, recipe, creator or a saved memory wins. Otherwise call recipe_pick: it returns the recipes that fit this project's format and footage, each with pickWhen, plus defaultId. Read the transcript, choose the ONE candidate whose pickWhen matches what is said (defaultId when none clearly fits), do its "before" step if any, then recipe_apply_style + recipe_render and follow the prompt: the recipe decides the style (captions, pacing, kinds of moves, music, transitions). Graphics inside it are designed by you: a named template is the reference look; author your own (motion_render_html, same palette and type) when a beat needs something the template can't show. (An app without recipe_pick, 2.0.5 and older: recipe_list with format short or long, and choose by description and footage.) Tell the user which recipe and why in one line and name one or two alternatives. Only when no candidate exists or the user asked for a plain clean-up: captions (caption_toggle, then caption_set_template: hormoziEmphasis for vertical, classic for long-form; not glowStack unless asked), a few motion graphics, emphasis zooms.
+7. The style, after steps 2-5 (they shift the timeline). The user's named style, recipe, creator or a saved memory wins. Otherwise call recipe_pick: it returns the recipes that fit this project's format and footage, each with pickWhen, plus defaultId. Read the transcript, choose the ONE candidate whose pickWhen matches what is said (defaultId when none clearly fits; a Short cut from a long-form video comes back fromLongForm and its defaultId is made for that), do its "before" step if any, then recipe_apply_style + recipe_render and follow the prompt: the recipe decides the style (captions, pacing, kinds of moves, music, transitions). Graphics inside it are designed by you: a named template is the reference look; author your own (motion_render_html, same palette and type) when a beat needs something the template can't show. (An app without recipe_pick, 2.0.5 and older: recipe_list with format short or long, and choose by description and footage.) Tell the user which recipe and why in one line and name one or two alternatives. Never trade the default for a plain edit because no candidate fits perfectly. Only when no candidate exists or the user asked for a plain clean-up: captions (caption_toggle, then caption_set_template: hormoziEmphasis for vertical, classic for long-form; not glowStack unless asked), a few motion graphics, emphasis zooms.
 8. B-roll, last: pick the moments yourself from what's said (any language), check them with project_check_broll, show the user your moments and ASK: their own images, generated on their image connector (only if one is connected; it uses their credits), or no B-roll. Never generate or invent images without their answer. Place what they give with project_add_broll. Nobody to ask: skip it and say so.
 9. llm_generate_title / description / timestamps, then preview_show and ask before export_start. Never export before the user has seen the preview.
 Report the real count each step returned. No music, intro/outro cards or FX overlays unless the recipe or the user asks for them.
@@ -8669,6 +8720,8 @@ Publishing has daily limits (YouTube uploads and updates per license, social pos
 
 PandaStudio problems (a capability gap like an imported font the export can't draw, a command failing unexpectedly, a render not matching the project): call system_report_issue yourself, then tell the user you reported it. Not for user mistakes, missing keys or taste.
 
+Views: in chat apps that show MCP views, preview_show, project_render_frame / _render_sheet and project_edit_summary also show the frames and what the edit contains, and clips_get / clips_make show the clips as cards the user can open, export or remove. Summarise in a line instead of describing each frame or clip; what the user does in a view arrives as context, don't redo it.
+
 Review items: the user can pin notes and drawings on the preview and send them in one turn (listed in the editor context and the message, crops attached in item order). Apply each one, then review_resolve with its id. review_list shows the open ones with a crop image path.
 
 Destination decides everything: infer it (Shorts/TikTok/vertical/<=90s: shorts; YouTube/tutorial/>8 min: youtube-long; LinkedIn; Loom/internal), else ask once.
@@ -8683,23 +8736,49 @@ const server = new Server(
 	{
 		capabilities: {
 			tools: {},
+			// Only the ui:// views (MCP Apps), listed for clients that render them.
+			resources: {},
 		},
 		instructions: INSTRUCTIONS,
 	},
 );
 
+/** Whether the connected client renders MCP Apps views (apps/index.mjs). */
+const uiEnabled = () => clientSupportsUi(server.getClientCapabilities(), { inApp: IN_APP });
+/** Image paths the app reported this session, served to views by ui_read_media. */
+const media = new MediaRegistry();
+
 server.setRequestHandler(ListToolsRequestSchema, async () => {
-	return {
-		tools: TOOLS.map((t) => ({
+	const ui = uiEnabled();
+	const tools = TOOLS.map((t) => {
+		const meta = ui ? toolUiMeta(t.name) : undefined;
+		return {
 			name: t.name,
 			description: t.description,
 			// Gemini rejects multi-type fields that JSON Schema allows.
 			inputSchema: portableSchema(t.inputSchema),
-		})),
-	};
+			...(meta ? { _meta: meta } : {}),
+		};
+	});
+	return { tools: ui ? [...tools, ...APP_TOOLS] : tools };
+});
+
+server.setRequestHandler(ListResourcesRequestSchema, async () => ({
+	resources: uiEnabled() ? listUiResources() : [],
+}));
+server.setRequestHandler(ListResourceTemplatesRequestSchema, async () => ({
+	resourceTemplates: [],
+}));
+server.setRequestHandler(ReadResourceRequestSchema, async (request) => {
+	const res = readUiResource(request.params.uri, mcpVersion());
+	if (!res) throw new Error(`unknown resource: ${request.params.uri}`);
+	return res;
 });
 
 server.setRequestHandler(CallToolRequestSchema, async (request, extra) => {
+	if (request.params.name === "ui_read_media") {
+		return media.read(request.params.arguments?.paths);
+	}
 	const tool = TOOLS.find((t) => t.name === request.params.name);
 	if (!tool) {
 		return {
@@ -8753,6 +8832,7 @@ server.setRequestHandler(CallToolRequestSchema, async (request, extra) => {
 		];
 		try {
 			const data = result?.data ?? result;
+			media.note(data);
 			const previews = collectInlinePreviewPaths(command, data);
 			for (const previewPath of previews) {
 				const fs = await import("node:fs/promises");

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.259.0",
+	"version": "1.262.0",
 	"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",