@writepanda/mcp

@writepanda/mcp

Model Context Protocol server for PandaStudio — a desktop video editor for YouTube creators. Lets AI agents edit videos like a human does: transcribe, delete filler words, drop motion graphics / FX / lower-thirds, generate captions, render the final MP4 — all without the user leaving their chat.

Works with Claude Desktop, Cursor, Continue.dev, Cline, and any MCP-compliant client.

You also need PandaStudio installed. The MCP server is a thin translator between MCP and PandaStudio's localhost-only automation API. Get the desktop app at writepanda.ai.

Install

The MCP server runs on demand via npx — no global install needed. Add to your client config:

Claude Desktop

~/Library/Application Support/Claude/claude_desktop_config.json (Mac) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "pandastudio": {
      "command": "npx",
      "args": ["-y", "@writepanda/mcp"]
    }
  }
}

Cursor

.cursor/mcp.json in your workspace:

{
  "mcpServers": {
    "pandastudio": {
      "command": "npx",
      "args": ["-y", "@writepanda/mcp"]
    }
  }
}

Continue.dev / Cline

Same shape, in their respective MCP config files.

After adding, restart your client. The MCP server auto-launches PandaStudio if it isn't running and waits for the localhost HTTP server to come up (~5s).

What the agent gets

55 tools covering the full PandaStudio editorial surface — complete UI parity:

Category	Tools
Discovery	system_status, system_list_commands
Project lifecycle	project_list, project_show, project_read, project_new, project_open, project_save, project_delete
Clips	project_add_clip, project_remove_clip, project_split_clip
Composition	project_add_motion_graphic, project_add_fx, project_add_lower_third, project_add_zoom, project_add_trim, project_add_speed, project_add_annotation
Region editing	project_remove_region, project_update_region
Canvas & style	project_set_aspect_ratio, project_set_wallpaper, project_set_style, project_set_crop, project_set_webcam_layout, project_set_export_settings
Transcript editing	transcript_transcribe, transcript_get, transcript_delete_words, transcript_remove_fillers, transcript_search, transcript_find_replace, transcript_remove_silences
Audio	audio_clean (DeepFilter denoising)
Captions	caption_toggle, caption_set_template, caption_set_style
Motion graphics	motion_list, motion_themes, motion_generate, motion_render_html
Assets	asset_list_sounds, asset_list_fx
AI metadata	llm_generate_title, llm_generate_description, llm_generate_timestamps
Export	export_start, export_list
Preview overlay	preview_show, preview_seek, preview_hide
Async jobs	job_wait, job_get
Escape hatch	pandastudio_call (raw verb.noun dispatch for anything not in the static list)

Idiomatic agent prompt

Edit the two videos in /Users/me/Downloads/raw-clips/ — transcribe them,
remove filler words, add a "How I Built This" intro card, enable bold-yellow
captions, and export the final MP4. Show me the preview overlay so I can
watch as you work.

The agent calls project_new, transcript_transcribe, transcript_remove_fillers, motion_generate, project_add_motion_graphic, caption_toggle, caption_set_template, preview_show, then export_start. ~12 tool calls, fully unattended.

How it works

MCP client (Cursor / Continue / Cline / Claude Desktop)
    ↓ JSON-RPC over stdio
@writepanda/mcp (this server)
    ↓ HTTP POST 127.0.0.1:7878/v1/call
PandaStudio desktop app
    ↓ in-process function calls
Editorial primitives (transcript, motion graphics, export, ...)

Every call is bearer-authenticated against the per-launch token PandaStudio writes to ~/.config/pandastudio/token (Mac/Linux) or %APPDATA%\pandastudio\token (Windows). Loopback-only — nothing leaves your machine.

Licensing

The MCP server honours the same license gate the desktop app does. With an expired trial and no license, only diagnostic tools work and every editorial tool returns { ok: false, details: { code: "trial_expired" } }. Activate a license in the desktop app's Settings → License panel.

Companions

• @writepanda/cli — same surface as a shell command (pandastudio system.status). Use this when scripting from bash / CI.
• Bundled Claude Skill — auto-loaded markdown instructions for Claude Code / Claude Desktop. Install via PandaStudio Settings → Local automation → Install Skill (drops into ~/.claude/skills/pandastudio/).

Documentation

• Full surface + recipes: writepanda.ai/cli
• Source: github.com/kamskans/openscreen

License

MIT.