ElevenLabs MCP

!export

<div class="title-block" style="text-align: center;" align="center">

   

</div>

<p align="center"> Official ElevenLabs <a href="https://github.com/modelcontextprotocol">Model Context Protocol (MCP)</a> server that enables interaction with powerful Text to Speech and audio processing APIs. This server allows MCP clients like <a href="https://www.anthropic.com/claude">Claude Desktop</a>, <a href="https://www.cursor.so">Cursor</a>, <a href="https://codeium.com/windsurf">Windsurf</a>, <a href="https://github.com/openai/openai-agents-python">OpenAI Agents</a> and others to generate speech, clone voices, transcribe audio, and more. </p>

<!-- mcp-name: io.github.elevenlabs/elevenlabs-mcp -->

Quickstart with Claude Desktop

1. Get your API key from ElevenLabs. There is a free tier with 10k credits per month.
2. Install uv (Python package manager), install with curl -LsSf https://astral.sh/uv/install.sh | sh or see the uv repo for additional install methods.
3. Go to Claude > Settings > Developer > Edit Config > claude_desktop_config.json to include the following:

{
  "mcpServers": {
    "ElevenLabs": {
      "command": "uvx",
      "args": ["elevenlabs-mcp"],
      "env": {
        "ELEVENLABS_API_KEY": "<insert-your-api-key-here>"
      }
    }
  }
}

If you're using Windows, you will have to enable "Developer Mode" in Claude Desktop to use the MCP server. Click "Help" in the hamburger menu at the top left and select "Enable Developer Mode".

Other MCP clients

For other clients like Cursor and Windsurf, run:

1. pip install elevenlabs-mcp
2. python -m elevenlabs_mcp --api-key={{PUT_YOUR_API_KEY_HERE}} --print to get the configuration. Paste it into appropriate configuration directory specified by your MCP client.

That's it. Your MCP client can now interact with ElevenLabs through these tools:

Example usage

⚠️ Warning: ElevenLabs credits are needed to use these tools.

Try asking Claude:

• "Create an AI agent that speaks like a film noir detective and can answer questions about classic movies"
• "Generate three voice variations for a wise, ancient dragon character, then I will choose my favorite voice to add to my voice library"
• "Convert this recording of my voice to sound like a medieval knight"
• "Create a soundscape of a thunderstorm in a dense jungle with animals reacting to the weather"
• "Turn this speech into text, identify different speakers, then convert it back using unique voices for each person"

Optional features

File Output Configuration

You can configure how the MCP server handles file outputs using these environment variables in your claude_desktop_config.json:

• ELEVENLABS_MCP_BASE_PATH: Specify the base path for file operations (default: ~/Desktop). This directory is also the security boundary for input files: any path passed to a tool that reads a local file (e.g. speech_to_text, isolate_audio, speech_to_speech, video_to_music, upload_music_for_inpainting) must resolve inside this directory, whether given as an absolute or relative path. Paths outside it — even if absolute and previously accepted — are rejected. Set this to a directory that contains everything you need to read from or write to.
• ELEVENLABS_MCP_OUTPUT_MODE: Control how generated files are returned (default: files)

Output Modes

The ELEVENLABS_MCP_OUTPUT_MODE environment variable supports three modes:

1. files (default): Save files to disk and return file paths

   "env": {
     "ELEVENLABS_API_KEY": "your-api-key",
     "ELEVENLABS_MCP_OUTPUT_MODE": "files"
   }

1. resources: Return files as MCP resources; binary data is base64-encoded, text is returned as UTF-8 text

   "env": {
     "ELEVENLABS_API_KEY": "your-api-key",
     "ELEVENLABS_MCP_OUTPUT_MODE": "resources"
   }

1. both: Save files to disk AND return as MCP resources

   "env": {
     "ELEVENLABS_API_KEY": "your-api-key",
     "ELEVENLABS_MCP_OUTPUT_MODE": "both"
   }

Resource Mode Benefits:

• Files are returned directly in the MCP response as base64-encoded data
• No disk I/O required - useful for containerized or serverless environments
• MCP clients can access file content immediately without file system access
• In both mode, resources can be fetched later using the elevenlabs://filename URI pattern

Use Cases:

• files: Traditional file-based workflows, local development
• resources: Cloud environments, MCP clients without file system access
• both: Maximum flexibility, caching, and resource sharing scenarios

Data residency keys

You can specify the data residency region with the ELEVENLABS_API_RESIDENCY environment variable. Defaults to "us".

Note: Data residency is an enterprise only feature. See the docs for more details.

Contributing

If you want to contribute or run from source:

1. Clone the repository:

git clone https://github.com/elevenlabs/elevenlabs-mcp
cd elevenlabs-mcp

1. Create a virtual environment and install dependencies using uv:

uv venv
source .venv/bin/activate
uv pip install -e ".[dev]"

1. Copy .env.example to .env and add your ElevenLabs API key:

cp .env.example .env
# Edit .env and add your API key

1. Run the tests to make sure everything is working:

./scripts/test.sh
# Or with options
./scripts/test.sh --verbose --fail-fast

1. Install the server in Claude Desktop: mcp install elevenlabs_mcp/server.py

1. Debug and test locally with MCP Inspector: mcp dev elevenlabs_mcp/server.py

Troubleshooting

Logs when running with Claude Desktop can be found at:

• Windows: %APPDATA%\Claude\logs\mcp-server-elevenlabs.log
• macOS: ~/Library/Logs/Claude/mcp-server-elevenlabs.log

Timeouts when using certain tools

Certain ElevenLabs API operations, like voice design and audio isolation, can take a long time to resolve. When using the MCP inspector in dev mode, you might get timeout errors despite the tool completing its intended task.

This shouldn't occur when using a client like Claude.

MCP ElevenLabs: spawn uvx ENOENT

If you encounter the error "MCP ElevenLabs: spawn uvx ENOENT", confirm its absolute path by running this command in your terminal:

which uvx

Once you obtain the absolute path (e.g., /usr/local/bin/uvx), update your configuration to use that path (e.g., "command": "/usr/local/bin/uvx"). This ensures that the correct executable is referenced.