@writepanda/mcp 1.55.0 → 1.57.0

package/bin/server.mjs

@@ -1193,7 +1193,7 @@ const TOOLS = [
 	{
 		name: "project_set_webcam_layout",
 		description:
-			"Set the webcam layout preset and optional position/scale for the current project. Use 'none' to hide the webcam overlay entirely.",
+			"Set the PROJECT-LEVEL webcam layout preset and optional position/scale. Use 'none' to hide the webcam. 'podcast' = two co-equal speaker tiles (host + guest). 'podcast-host-full'/'podcast-guest-full' = a single speaker full-frame (host=mediaPath/speaker 1, guest=webcamPath/speaker 2). To vary the layout PER SECTION of a podcast, split the clip and use project_set_clip_layout instead.",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -1202,7 +1202,7 @@ const TOOLS = [
 				preset: {
 					type: "string",
 					description:
-						"picture-in-picture | vertical-stack | side-by-side | none. 'none' hides the webcam.",
+						"picture-in-picture | vertical-stack | side-by-side | podcast | podcast-host-full | podcast-guest-full | none",
 				},
 				cx: {
 					type: "number",
@@ -1443,20 +1443,61 @@ const TOOLS = [
 	{
 		name: "project_set_clip_kind",
 		description:
-			"Stamp a clip's capture origin authoritatively: kind = camera | screen | upload. project.read returns each clip's kind (your visual-strategy signal: camera → designed segments + emphasis zooms; screen → cursor zooms). On pre-v1.28 projects kind wasn't recorded, so project.read INFERS it (flagged kindInferred:true) — and inference can be wrong. Use this to correct/lock it. `camera` = talking-head; `screen` = screen recording (maybe with webcam PiP); `upload` = imported file.",
+			"Stamp a clip's capture origin authoritatively: kind = camera | screen | upload | podcast. project.read returns each clip's kind (your visual-strategy signal: camera → designed segments + emphasis zooms; screen → cursor zooms). On pre-v1.28 projects kind wasn't recorded, so project.read INFERS it (flagged kindInferred:true) — and inference can be wrong. Use this to correct/lock it. `camera` = talking-head; `screen` = screen recording (maybe with webcam PiP); `upload` = imported file; `podcast` = two-speaker composite (host=mediaPath, guest=webcamPath).",
 		inputSchema: {
 			type: "object",
 			properties: {
 				id: { type: "string" },
 				path: { type: "string" },
 				clipId: { type: "string", description: "Clip id from project_read mainTrack.clips[n].id" },
-				kind: { type: "string", enum: ["camera", "screen", "upload"] },
+				kind: { type: "string", enum: ["camera", "screen", "upload", "podcast"] },
 				expectedRevision: { type: "number" },
 			},
 			required: ["clipId", "kind"],
 		},
 		command: "project.set-clip-kind",
 	},
+	{
+		name: "project_set_webcam_offset",
+		description:
+			"Fine-tune a podcast composite's guest/host sync. offsetMs shifts the guest (webcamPath) relative to the host: positive delays the guest, negative advances it, 0 clears it (±10000). Honored in preview + export — the guest VIDEO and AUDIO shift together. Use when the two speakers look or sound slightly out of sync.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				clipId: { type: "string", description: "Clip id of the podcast composite" },
+				offsetMs: {
+					type: "number",
+					description: "Guest time nudge in ms (±10000). Positive delays guest; 0 clears.",
+				},
+				expectedRevision: { type: "number" },
+			},
+			required: ["clipId", "offsetMs"],
+		},
+		command: "project.set-webcam-offset",
+	},
+	{
+		name: "project_set_clip_layout",
+		description:
+			"Set ONE clip's webcam layout — the per-section override for a podcast. preset: podcast | podcast-host-full | podcast-guest-full | picture-in-picture | side-by-side | vertical-stack | none (none clears the override so the clip inherits the project layout). 'podcast-host-full' shows only speaker 1 (host=mediaPath) full-frame; 'podcast-guest-full' shows only speaker 2 (guest=webcamPath). Speaker-driven editing flow: transcript_get returns each word's `speaker` ('host'/'guest') for podcast clips; project_split_clip at the points where the active speaker changes; then call this per section to cut to whoever is talking. Honored in preview + export.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				clipId: { type: "string", description: "Clip id of the section to set" },
+				preset: {
+					type: "string",
+					description:
+						"podcast | podcast-host-full | podcast-guest-full | picture-in-picture | side-by-side | vertical-stack | none",
+				},
+				expectedRevision: { type: "number" },
+			},
+			required: ["clipId", "preset"],
+		},
+		command: "project.set-clip-layout",
+	},
 
 	// ── transcript-based editing ────────────────────────────────────
 	{
@@ -1476,7 +1517,7 @@ const TOOLS = [
 	{
 		name: "transcript_get",
 		description:
-			"Return the merged edited-time transcript: every word with id, text, startMs, endMs in EDITED-TIMELINE coordinates. Use word IDs as input to transcript_delete_words.",
+			"Return the merged edited-time transcript: every word with id, text, startMs, endMs in EDITED-TIMELINE coordinates. Use word IDs as input to transcript_delete_words. For a PODCAST composite, each word also carries `speaker` ('host' = speaker 1 / mediaPath, 'guest' = speaker 2 / webcamPath) — use these spans to drive project_set_clip_layout (cut to whoever is talking).",
 		inputSchema: {
 			type: "object",
 			properties: {
@@ -2230,7 +2271,7 @@ const TOOLS = [
 const server = new Server(
 	{
 		name: "pandastudio",
-		version: "1.55.0",
+		version: "1.57.0",
 	},
 	{
 		capabilities: {

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.55.0",
+	"version": "1.57.0",
 	"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",