limostudio-mcp

LimoStudio for your AI app. Add this MCP server to Claude Code, Claude Desktop, Cursor, VS Code, Codex, Gemini CLI or another app that runs MCP servers, and ask it for a video: it opens a LimoStudio window you can watch and edits your project there, with the same 100+ tools as LimoStudio's own assistant — clips, text and captions, scenes and versions, color, effects and keyframes, music and synths, mixing, mastering, AI images, video and songs, and export.

Every change is an undo step in the studio, your AI's calls run one at a time in the order it sends them, and anything that spends LimoStudio credits (an AI generation) asks you first, in the studio window.

Full guide: https://limostudio.ai/mcp**

What you need

• A LimoStudio account (free, with trial credits)
• Google Chrome (or Microsoft Edge — see the settings below)
• Node.js 20 or newer (it brings npx)
• An AI app that runs MCP servers on your computer

Add it to your AI app

Add these lines, then restart the app. Already have other servers there? Put the limostudio entry next to them. Each one runs the latest server, so there is nothing to update by hand.

Claude Code

Run in a terminal:

claude mcp add -s user limostudio -- npx -y limostudio-mcp@latest

Claude Desktop

Add to claude_desktop_config.json — Settings › Developer › Edit Config opens it:

{
  "mcpServers": {
    "limostudio": {
      "command": "npx",
      "args": ["-y", "limostudio-mcp@latest"]
    }
  }
}

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "limostudio": {
      "command": "npx",
      "args": ["-y", "limostudio-mcp@latest"]
    }
  }
}

VS Code

Add to mcp.json — Command Palette › MCP: Open User Configuration opens it:

{
  "servers": {
    "limostudio": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "limostudio-mcp@latest"]
    }
  }
}

Codex

Add to ~/.codex/config.toml:

[mcp_servers.limostudio]
command = "npx"
args = ["-y", "limostudio-mcp@latest"]
startup_timeout_sec = 60
tool_timeout_sec = 600

Gemini CLI

Run in a terminal:

gemini mcp add -s user limostudio npx -y limostudio-mcp@latest

Other apps

Most other MCP apps take the same lines in their MCP settings:

{
  "mcpServers": {
    "limostudio": {
      "command": "npx",
      "args": ["-y", "limostudio-mcp@latest"]
    }
  }
}

The first time

1. Ask your AI for something small: "Open LimoStudio and add a title that says Hello."
2. A LimoStudio window opens in Chrome, with a browser profile kept for the AI (~/.limostudio/agent-chrome).
3. Your everyday browser opens LimoStudio's sign-in (Google, Apple or an email code); then the window is signed in too, and stays so. (Google doesn't allow signing in inside a browser a program drives, so the sign-in is handed over from your own browser.)
4. The studio asks: "An AI agent (Claude Code) wants to control this studio…" — Allow. The header then shows the agent and its latest step; click it to disconnect.

You stay in charge

• An AI connects only to the LimoStudio window it opened itself; your everyday browser never takes agent connections.
• The studio asks before an AI connects — in that window's Settings › AI Assistant › AI agent connection (the window keeps its own settings): ask each time (default), allow on this device, or never allow.
• Anything that spends credits asks again, each time, unless you allow it in the same settings.
• Every change the AI makes is one undo step, and its calls run one at a time, in the order it sends them.

Tools

LimoStudio's assistant tools (read from https://limostudio.ai/api/agent-tools each time the server starts, so new ones arrive without an update), plus the server's own: open_studio, screenshot (the AI's eyes — time: 12 seeks there first and waits for that moment's frame), save_project, studio_info (the open project, the tools waiting for your yes in the window, the files the window downloaded to this computer — an export's path), undo and disconnect.

An AI that cannot see or hear the result has tools for that too: transcribe_clip hears a clip (and scores a shot's line against its dialogue), query_mixer loudness:true measures every track, export_status follows an export to its file, wait_shots waits for a batch of shots. set_ai_model sets the model picks and the assistant's defaults — the free ComfyUI models run on your own ComfyUI, set up in the studio window the AI works in (Settings › ComfyUI).

Settings

Environment variables for the server (in your app's MCP settings, "env": { … }):

Troubleshooting

• Chrome isn't found — install it, or set LIMOSTUDIO_BROWSER=msedge.
• "npx not found" — some apps don't see a Node.js installed with a version manager: put the full path of npx in "command" (which npx shows it).
• The first request timed out — it waited while you signed in or allowed it. Ask again; the window stays signed in.
• Another account — sign out in the window, or delete ~/.limostudio/agent-chrome.

More: https://limostudio.ai/mcp

License

MIT © Frinet. LimoStudio is a trademark of Frinet; using LimoStudio needs an account and follows its terms.