@writepanda/mcp 1.105.0 → 1.109.0

package/bin/server.mjs

@@ -1287,7 +1287,7 @@ const TOOLS = [
 				outline: {
 					type: "boolean",
 					description:
-						"Draw a person OUTLINE (colored keyline + drop shadow) around the cutout. Enable on 'remove' over a cream wallpaper for the VOX magazine-cutout look; also composes with 'blur'.",
+						"Person OUTLINE (colored keyline + drop shadow) around the cutout — the VOX magazine-cutout look. DEFAULTS ON for mode 'remove'; pass outline:false to remove the background with no keyline. For 'blur' it's off unless set true.",
 				},
 				outlineWidth: {
 					type: "number",
@@ -1299,7 +1299,8 @@ const TOOLS = [
 				},
 				outlineShadow: {
 					type: "boolean",
-					description: "Whether the outline casts a drop shadow (lifted-sticker look). Default true.",
+					description:
+						"Whether the outline casts a drop shadow (lifted-sticker look). Default true.",
 				},
 				matteContract: {
 					type: "number",
@@ -1925,6 +1926,46 @@ const TOOLS = [
 		},
 		command: "project.set-focal-point",
 	},
+	{
+		name: "project_auto_reframe",
+		description:
+			"ACTIVE-SPEAKER AUTO-REFRAME — a tracked virtual camera (Opus-Clip style) for director-cut footage (talk shows, interviews, multi-cam) reframed into a vertical/square canvas. Detects shot cuts, then densely tracks faces (MediaPipe, bundled offline, with tiling for small/far faces) and, on multi-person shots, frames WHOEVER IS TALKING (mouth-open × audio speech-energy) rather than the biggest face. The crop smoothly PANS to follow the subject within a shot and CUTS at shot boundaries, with a safe-zone clamp so the face never leaves frame. This is the right tool (not project_set_focal_point) when a landscape source with multiple/alternating people is cut to 9:16 — set-focal-point sets one static point for the whole clip. Async: opens a renderer for the dense pass (allow a couple minutes for a few-minute source). Optional zoom = fixed punch-in (omit for adaptive). Skips clips already matching the canvas aspect. clear=true reverts to the static crop. Preview and export render the crop identically, per frame.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				clipId: {
+					type: "string",
+					description:
+						"Target one clip (from project.read clipStates). Omit to reframe every landscape clip.",
+				},
+				threshold: {
+					type: "number",
+					description: "Scene-cut sensitivity 0-1 (default 0.3). Lower = more cuts.",
+				},
+				minShotMs: {
+					type: "number",
+					description: "Minimum shot length ms (default 500); closer cuts are merged.",
+				},
+				samplesPerShot: {
+					type: "number",
+					description: "Frames sampled per shot for the focal median (default 3).",
+				},
+				zoom: {
+					type: "number",
+					description:
+						"Fixed punch-in (≥1) for every shot. Omit for the default ADAPTIVE zoom that sizes each speaker's face to a consistent fraction of the frame. Pass e.g. 1.3 to force a uniform punch-in.",
+				},
+				clear: {
+					type: "boolean",
+					description: "Remove the auto-reframe track(s), reverting to the static cover-crop.",
+				},
+				expectedRevision: { type: "number" },
+			},
+		},
+		command: "project.auto-reframe",
+	},
 	{
 		name: "project_set_export_settings",
 		description:
@@ -2257,6 +2298,34 @@ const TOOLS = [
 		},
 		command: "transcript.delete-words",
 	},
+	{
+		name: "transcript_insert_words",
+		description:
+			"Add word(s) the transcription MISSED — the counterpart to transcript_find_replace (which only rewrites words that already exist). Speech-to-text sometimes drops a spoken word; this inserts it with plausible timing (fills the gap the drop left, sized to the local speaking rate; borrows a few ms from the previous word when neighbours are flush). Anchor with EITHER afterWordId (insert right after that word — usual) OR beforeWordId (insert right before it — to add at the very start). Word ids come from transcript_get. `text` may be multiple space-separated words. Non-destructive — existing word ids are untouched, so trims/zooms/captions anchored elsewhere are unaffected. Does NOT re-transcribe.",
+		inputSchema: {
+			type: "object",
+			properties: {
+				id: { type: "string" },
+				path: { type: "string" },
+				afterWordId: {
+					type: "string",
+					description: "Insert immediately AFTER this word id (from transcript_get).",
+				},
+				beforeWordId: {
+					type: "string",
+					description:
+						"Insert immediately BEFORE this word id (use to add at the very start). Provide exactly one of afterWordId / beforeWordId.",
+				},
+				text: {
+					type: "string",
+					description: "The word(s) to insert. Space-separated text inserts multiple words.",
+				},
+				expectedRevision: { type: "number" },
+			},
+			required: ["text"],
+		},
+		command: "transcript.insert-words",
+	},
 	{
 		name: "transcript_restore_words",
 		description:
@@ -2474,6 +2543,11 @@ const TOOLS = [
 					description:
 						"Font size in REM, e.g. '2.6rem'. Clamped to 1.0–5.0rem (the editor's slider range).",
 				},
+				uppercase: {
+					type: "boolean",
+					description:
+						"Force ALL-CAPS caption text. true uppercases every caption word; false restores the template's own casing (use false to turn OFF a template that ships uppercase, e.g. 'editorial').",
+				},
 				expectedRevision: { type: "number" },
 			},
 		},

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@writepanda/mcp",
-	"version": "1.105.0",
+	"version": "1.109.0",
 	"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",