The Hume MCP Server lets you run the Octave Text-to-Speech (TTS) tools inside MCP-enabled clients. This server exposes a set of MCP tools that you can invoke from an AI chat agent or integrated client to synthesize speech, manage voices, and play or reuse previously generated audio. Core tools include tts for generating spoken audio from text, list_voices to fetch available voices, save_voice to store a voice in your library, delete_voice to remove voices, and play_previous_audio to replay recently generated audio. To use it, configure your MCP client to point to the Hume MCP Server entry (for example via npx as shown in the Quickstart), provide your Hume API key, and then call the tools by name with the appropriate arguments as part of your conversation prompts or agent actions. The server is designed to integrate with clients like Claude Desktop, Cursor, or Windsurf, enabling rich TTS narration within AI-assisted workflows.

Prerequisites:\n- An account and API Key from Hume AI (HUME_API_KEY)\n- Node.js installed on your machine (client-side MCP configuration uses npx)\n- A supported MCP client (e.g., Claude Desktop, Cursor, Windsurf) to connect to the MCP server.\n\n1) Install Node.js from https://nodejs.org/ (ensure node and npm are in your PATH).\n2) Obtain your Hume API Key from the Hume AI dashboard.\n3) Start the MCP server via npx using the provided configuration. This will fetch the @humeai/mcp-server package and run it in the MCP client context. Example:\n\n```

In your MCP client configuration file (e.g., claude_desktop_config.json or mcp.json for Cursor), use:\n\n{"hume": {\n "command": "npx",\n "args": ["-y", "@humeai/mcp-server"],\n "env": {"HUME_API_KEY": "<your_hume_api_key>"}\n}}\n```

\n4) Replace <your_hume_api_key> with the API key from your Hume account.\n5) If desired, set optional environment variables for local working directories or modes as described in the server docs.\n6) Verify that the MCP client can communicate with the Hume MCP Server and begin issuing MCP tool commands.

Environment variable tips: - HUME_API_KEY is required to authorize requests. - WORKDIR can be set to define where audio files are stored; defaults to the system temp directory plus /hume-tts. - EMBEDDED_AUDIO_MODE toggles whether audio is embedded in responses. - ANTHROPIC_API_KEY is only needed if you run the evaluation framework. Common issues often involve misconfigured API keys or incorrect client configuration. Ensure your MCP client is using the same server name (e.g., hume) and that the npx invocation resolves to the correct package version. If you encounter audio playback issues, verify that a compatible audio player (like ffplay) is installed and accessible in your system PATH.