voxtral-mcp

Local voice for any MCP client (Claude Code, Claude Desktop, Cursor, etc.) via Mistral Voxtral 4B TTS (MLX 4-bit). Higher voice quality than the lightweight pocket-tts route, at the cost of a larger model and slower TTFA.

• 9 languages: 🇬🇧 English, 🇫🇷 French, 🇩🇪 German, 🇪🇸 Spanish, 🇮🇹 Italian, 🇵🇹 Portuguese, 🇳🇱 Dutch, 🇮🇳 Hindi, 🇸🇦 Arabic
• 4 B parameters, 4-bit MLX quantization (~2.5 GB on disk)
• ~2.4× real-time generation on Apple Silicon M-series
• TTFA ~2 s thanks to native streaming via mlx-audio stream=True
• Non-blocking speak(), gap-free playback via sounddevice write-mode
• ~3 GB resident RAM once the model is loaded

⚠️ Licence: the Voxtral model itself is distributed by Mistral under CC BY-NC 4.0 — non-commercial use only. This wrapper's code is MIT.

Requirements

• macOS Apple Silicon (M1/M2/M3/M4) — required, MLX doesn't run on Intel
• ≥16 GB RAM recommended (the 4-bit model keeps ~3 GB resident)
• Python 3.10 – 3.14

Install

uvx voxtral-mcp --help

Or persistent:

uv tool install voxtral-mcp

Then add to your MCP client's .mcp.json:

{
  "mcpServers": {
    "voxtral": {
      "command": "uvx",
      "args": ["voxtral-mcp"]
    }
  }
}

MCP tools

Configuration

All env vars (in the env block of .mcp.json):

Voices

Voxtral ships with 20 preset voices across 9 languages. A non-exhaustive sample:

Pass voice="fr_female" in your speak() call to switch.

Claude Code users

If you're on Claude Code (CLI, desktop, or via Cursor), install the bundled plugin (MCP wiring + /voice-mode skill) in one shot:

/plugin marketplace add Vincweb/voxtral-mcp
/plugin install voxtral@vincweb-tools

See the main repo for architecture details and a side-by-side comparison with kyutai-tts-mcp (smaller / faster TTFA / permissive licence).

License

MIT for this wrapper. The underlying Voxtral model is CC BY-NC 4.0 (non-commercial) — see Mistral.