Overview

Cartesia MCP Server implements the Model Context Protocol (MCP) adapter that lets local agents and desktop assistants (for example Claude Desktop, Cursor, and other MCP-capable clients) call Cartesia’s audio models. Use the server to convert text to speech (TTS), localize existing speech into another language/voice, infill missing audio between clips, and otherwise manage voice transformations using Cartesia’s API while keeping tooling local and agent-friendly.

This server is intended for developers who want an easy, MCP-compatible bridge to Cartesia from an agent or IDE environment. It runs as an executable exposed to your MCP client, takes an API key for Cartesia, and accepts commands and file paths so downstream agents can request audio operations programmatically.

Features

Text-to-speech (TTS) using Cartesia voices
Voice localization: convert speech from one language/voice to another
Audio infill: generate audio to fill gaps between two clips
Voice replacement: transform an existing audio clip to use a different voice
MCP-compatible: integrates with Claude Desktop, Cursor, and other MCP clients
Local file output support with configurable output directory

Installation / Configuration

Install the package from PyPI and verify the executable path:

pip install cartesia-mcp
which cartesia-mcp   # returns absolute path to the executable used in MCP configs

Create a Cartesia account and API key at https://play.cartesia.ai/sign-in. There is a free tier and you can create a key under API Keys → New.

Environment variables and configuration:

# Required environment variable:
export CARTESIA_API_KEY="sk-...your-api-key..."

# Optional: where generated audio files will be saved
export OUTPUT_DIRECTORY="/path/to/cartesia/output"

Typical environment/config keys:

Key	Required	Description
CARTESIA_API_KEY	Yes	Your Cartesia API key for server requests
OUTPUT_DIRECTORY	No	Local folder to store generated audio files (defaults to working directory)

Claude Desktop Integration

Edit Claude Desktop’s configuration file (Settings → Developer → Edit Config) and add an MCP server entry that points to the cartesia-mcp executable:

{
  "mcpServers": {
    "cartesia-mcp": {
      "command": "/absolute/path/to/cartesia-mcp",
      "env": {
        "CARTESIA_API_KEY": "sk-xxxx",
        "OUTPUT_DIRECTORY": "/Users/you/cartesia-output"
      }
    }
  }
}

Notes:

Use the absolute path returned by which cartesia-mcp.
Configure OUTPUT_DIRECTORY if you want all outputs in a specific folder.
When invoking infill or localization from Claude, provide absolute file paths to existing audio files.

Cursor Integration

Create either a project-level file .cursor/mcp.json or a global ~/.cursor/mcp.json:

{
  "cartesia-mcp": {
    "command": "/absolute/path/to/cartesia-mcp",
    "env": {
      "CARTESIA_API_KEY": "sk-xxxx",
      "OUTPUT_DIRECTORY": "/Users/you/cartesia-output"
    }
  }
}

Cursor will then present the Cartesia tools to your workspace for agent workflows.

Available Tools / Resources

The MCP server exposes a set of tools (actions) that MCP clients can call. Typical tool names and capabilities you will see in the MCP interface:

list_voices — return available Cartesia voices and metadata
text_to_speech — synthesize audio from a text prompt and voice selection
localize_speech — translate/convert an audio file into another language/voice
infill_audio — generate audio between two clips, given start/end audio
change_voice — re-render an existing clip in a different voice

Refer to the GitHub repository for the definitive list and current behavior: https://github.com/cartesia-ai/cartesia-mcp

Use Cases (Concrete Examples)

Text-to-Speech (from Claude or Cursor)
- Ask your agent: “Convert ‘Welcome to our app’ to audio using voice ‘amy’.” The agent will call text_to_speech and return a saved audio file path (e.g., /Users/you/cartesia-output/welcome_amy.wav).
Localize speech to another language
- Provide an existing clip path and request localization: “Localize /tmp/input_en.wav to Spanish using voice ‘sofia’.” The tool runs localize_speech and returns the output file path and metadata (language confidence, duration).
Infill missing audio between clips
- You have two clips, start.wav and end.wav. Ask the agent: “Create a smooth infill between these two files.” Provide absolute paths and a target voice. The server will call infill_audio and produce a new audio file containing the generated segment.
Replace voice on an existing clip
- Request: “Render /tmp/announcement.wav using voice ‘omar’.” This uses change_voice to create a voice-transformed copy.

Tips:

Always use absolute file paths when calling localize, infill, or change_voice so the MCP server can access files.
Inspect the output directory for generated assets; filenames usually include the operation and voice for clarity.

Troubleshooting & Next Steps

If the executable is not found, ensure your Python environment’s bin directory is on PATH or use the absolute path in the MCP config.
For permission errors, confirm the OUTPUT_DIRECTORY exists and is writable by the user running the MCP client.
For feature updates or issues, open an issue or review examples in the repository: https://github.com/cartesia-ai/cartesia-mcp

This MCP server streamlines integrating Cartesia audio capabilities into agent-driven workflows, enabling programmatic TTS, localization, and intelligent audio editing directly from desktop agents and developer tools.

Cartesia MCP Server for Voice Infill & TTS