pa-elevenlabs-mcp

A self-hosted HTTP MCP server wrapping the ElevenLabs speech-to-text API. Designed to be used with Claude Cowork plugins where the sandbox blocks direct outbound API calls. The MCP server runs on the host machine (outside the sandbox), so the API call goes through cleanly.

Tools exposed

Related MCP server: io.github.chicogong/ffvoice

Local development with Docker Compose + ngrok

1. Set up environment variables

cp .env.example .env

Edit .env and fill in:

ELEVENLABS_API_KEY=sk_...
NGROK_AUTHTOKEN=your_ngrok_authtoken   # get free token at https://ngrok.com

2. Start the server

docker compose up --build

This starts:

• mcp-server on port 8000

• ngrok tunnel on port 4040 (inspect UI)

3. Get your ngrok public URL

curl -s http://localhost:4040/api/tunnels | python3 -m json.tool | grep public_url

Or open http://localhost:4040 in your browser.

You'll get a URL like: https://abc123.ngrok-free.app

4. Test the health endpoint

curl https://abc123.ngrok-free.app/health
# {"status":"ok"}

Connect to a Cowork plugin

Add the following to your plugin's .mcp.json (or to claude_desktop_config.json for Claude Code):

{
  "mcpServers": {
    "elevenlabs-mcp": {
      "type": "http",
      "url": "https://abc123.ngrok-free.app/mcp"
    }
  }
}

Replace abc123.ngrok-free.app with your actual ngrok URL.

In your skill, call the tool like:

Use the elevenlabs-mcp speech_to_text tool with:
- file_path: /absolute/path/to/audio.mp3
- model_id: scribe_v2
- diarize: true

EC2 production deployment

For production, use the same docker-compose.yml but:

1. Remove the ngrok service entirely

2. Point your domain to port 8000 using nginx or Caddy as a reverse proxy

3. Use HTTPS (Let's Encrypt via Caddy is the easiest option)

Minimal Caddyfile:

your-domain.com {
    reverse_proxy localhost:8000
}

Update .mcp.json to use your real domain:

{
  "mcpServers": {
    "elevenlabs-mcp": {
      "type": "http",
      "url": "https://your-domain.com/mcp"
    }
  }
}