mcp-tts

mcp-tts is an MCP server that exposes multiple text-to-speech capabilities through the MCP protocol. It registers four tools you can call via MCP: say_tts (macOS built-in say), elevenlabs_tts (ElevenLabs API voices), google_tts (Google Gemini TTS models), and openai_tts (OpenAI TTS with various voices). You can run mcp-tts directly as a server and configure environment variables or CLI flags to control behavior such as concurrency, output saving, and speech suppression. Tools support different voice options, quality models, and rate/speed controls, allowing you to tailor audio output to your needs. The server is designed to integrate with Claude Desktop, Codex CLI, Gemini CLI, or any MCP-enabled client. To use the tools, ensure you have the necessary API keys for cloud-based services and optionally set environment variables to tweak behavior like concurrency, output saving, and spoken output formatting.

Prerequisites:

• Go installed on your system (for building or fetching the mcp-tts binary).
• Access to the internet to download the binary via go install or prebuilt releases.

Install steps:

1. Install the MCP TTS server:

go install github.com/blacktop/mcp-tts@latest

2. Verify installation:

mcp-tts --help

3. Configure environment variables (see README for details):

export ELEVENLABS_API_KEY=your_key
export ELEVENLABS_VOICE_ID=your_voice_id
export GOOGLE_AI_API_KEY=your_key
export OPENAI_API_KEY=your_key
export OPENAI_TTS_INSTRUCTIONS="Speak in a cheerful and positive tone"
export MCP_TTS_OUTPUT_DIR=/path/to/audio
export MCP_TTS_NO_PLAY=false

4. Run the server:

mcp-tts

5. Connect via MCP-using clients or add to Claude/Codex/Gemini settings as shown in the README.

Notes and tips:

• You can enable concurrent TTS by setting MCP_TTS_ALLOW_CONCURRENT to true, or disable it to enforce sequential playback for all requests.
• Suppress the default 'Speaking:' output by setting MCP_TTS_SUPPRESS_SPEAKING_OUTPUT to true or using the --suppress-speaking-output flag.
• Saving audio to disk requires enabling MCP_TTS_OUTPUT_DIR (and optionally MCP_TTS_NO_PLAY to skip playback). Files are saved with unique names like tts_{timestamp}_{hash}.{ext}.
• Each provider has its own configuration requirements (e.g., macOS say_tts is macOS-only; cloud-based providers require valid API keys and may incur costs).
• When using concurrent TTS, be mindful of potential overlapping audio. This feature is intended for advanced workflows where simultaneous speech is required.
• If you’re integrating with Claude/Code/Gemini clients, ensure the correct environment variables are passed to mcp-tts in your MCP server configuration.