The Cartesia MCP server provides a bridge for clients such as Cursor, Claude Desktop, and OpenAI agents to access Cartesia's API. It supports generating speech from text, localizing speech into different languages, and infilling audio between existing segments. The server can be installed via pip and run as a command-line executable. Clients configure the MCP server through integration files (Claude Desktop uses claude_desktop_config.json; Cursor uses .cursor/mcp.json or a global config). To use, you must have a Cartesia account with API keys; you can obtain an API key from the Cartesia Playground API Keys section (New). When running, you can specify OUTPUT_DIRECTORY to store generated files. The Claude Desktop example shows how to set environment variables including CARTESIA_API_KEY and OUTPUT_DIRECTORY. The README also mentions a free tier with 20,000 credits per month. The MCP server exposes commands to list voices, convert text to audio, localize speech, infill audio, and switch voices.