cartesia-mcp 0.1.8

The official Cartesia MCP server

Cartesia MCP Server

The Cartesia MCP server exposes Cartesia APIs over the Model Context Protocol (MCP) so clients such as Cursor, Claude Desktop, and OpenAI Agents can list voices, run TTS, clone voices, infill audio, and more—without one-off scripts.

Documentation: Cartesia docs — MCP (overview and links back to this README).

Requirements

• Python 3.13+ (required; enforced in pyproject.toml)
• A Cartesia API key — create one under API Keys after signing in at the playground. Free tier includes 20,000 credits per month.

Installation

pip

pip install cartesia-mcp
which cartesia-mcp   # copy absolute path for MCP config below

uv (recommended for ephemeral runs)

If you use uv, you can run the published package without a global install:

uvx cartesia-mcp

Use which uvx or command -v uvx if your MCP client needs an absolute path to uvx. Some clients invoke MCP via command + args (see Cursor).

Environment

Variable	Required	Description
CARTESIA_API_KEY	Yes	Your Cartesia API key
OUTPUT_DIRECTORY	No	Directory for generated audio files (defaults to .)

Claude Desktop integration

Edit Settings → Developer → Edit Config (claude_desktop_config.json).

Use either the installed executable (from pip install) or uvx:

Installed cartesia-mcp

{
  "mcpServers": {
    "cartesia-mcp": {
      "command": "/absolute/path/to/cartesia-mcp",
      "env": {
        "CARTESIA_API_KEY": "<your-api-key>",
        "OUTPUT_DIRECTORY": "/absolute/path/to/output"
      }
    }
  }
}

Omit OUTPUT_DIRECTORY to write files to the process working directory.

uvx

{
  "mcpServers": {
    "cartesia-mcp": {
      "command": "uvx",
      "args": ["cartesia-mcp"],
      "env": {
        "CARTESIA_API_KEY": "<your-api-key>"
      }
    }
  }
}

Try asking Claude to…

• List all available Cartesia voices
• Convert text to audio with a chosen voice
• Localize an existing voice into another language
• Infill audio between two segments (provide absolute paths to audio files)
• Change an audio file to use a different voice

Cursor integration

Create .cursor/mcp.json in your project or ~/.cursor/mcp.json globally. You can use the same shapes as Claude Desktop (command pointing at the installed binary, or uvx + args).

Tools

The server registers MCP tools for TTS (text_to_speech), voice listing and management (list_voices, get_voice, clone_voice, etc.), infill, voice change, and localize voice. See cartesia_mcp/server.py for the authoritative list and parameters.