Install via: uvx voice-mode | pip install voice-mode | getvoicemode.com

Natural voice conversations for AI assistants. Voice Mode brings human-like voice interactions to Claude, ChatGPT, and other LLMs through the Model Context Protocol (MCP).

Runs on: Linux • macOS • Windows (WSL) | Python: 3.10+ | Tested: Ubuntu 24.04 LTS, Fedora 42

• 🎙️ Voice conversations with Claude - ask questions and hear responses
• 🔄 Multiple transports - local microphone or LiveKit room-based communication
• 🗣️ OpenAI-compatible - works with any STT/TTS service (local or cloud)
• ⚡ Real-time - low-latency voice interactions with automatic transport selection
• 🔧 MCP Integration - seamless with Claude Desktop and other MCP clients

All you need to get started:

1. 🔑 OpenAI API Key (or compatible service) - for speech-to-text and text-to-speech
2. 🎤 Computer with microphone and speakers OR ☁️ LiveKit server (LiveKit Cloud or self-hosted)

claude mcp add --scope user voice-mode uvx voice-mode
export OPENAI_API_KEY=your-openai-key
claude
> /converse

Watch Voice Mode in action:

Once configured, try these prompts with Claude:

• "Let's have a voice conversation"
• "Ask me about my day using voice"
• "Tell me a joke" (Claude will speak and wait for your response)
• "Say goodbye" (Claude will speak without waiting)

The new converse function makes voice interactions more natural - it automatically waits for your response by default.

Add to your Claude Desktop configuration file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "voicemode": {
      "command": "uvx",
      "args": ["voice-mode"],
      "env": {
        "OPENAI_API_KEY": "your-openai-key"
      }
    }
  }
}

{
  "mcpServers": {
    "voicemode": {
      "command": "voicemode",
      "env": {
        "OPENAI_API_KEY": "your-openai-key"
      }
    }
  }
}

Note: The converse tool is the primary interface for voice interactions, combining speaking and listening in a natural flow.

📖 See docs/configuration.md for complete setup instructions for all MCP hosts

📁 Ready-to-use config files in config-examples/

The only required configuration is your OpenAI API key:

export OPENAI_API_KEY="your-key"

# Custom STT/TTS services (OpenAI-compatible)
export STT_BASE_URL="http://localhost:2022/v1"  # Local Whisper
export TTS_BASE_URL="http://localhost:8880/v1"  # Local TTS
export TTS_VOICE="alloy"                        # Voice selection

# LiveKit (for room-based communication)
# See docs/livekit/ for setup guide
export LIVEKIT_URL="wss://your-app.livekit.cloud"
export LIVEKIT_API_KEY="your-api-key"
export LIVEKIT_API_SECRET="your-api-secret"

# Debug mode
export VOICE_MCP_DEBUG="true"

# Save all audio (TTS output and STT input)
export VOICE_MCP_SAVE_AUDIO="true"

For privacy-focused or offline usage, Voice Mode supports local speech services:

• Whisper.cpp - Local speech-to-text with OpenAI-compatible API
• Kokoro - Local text-to-speech with multiple voice options

These services provide the same API interface as OpenAI, allowing seamless switching between cloud and local processing.

By strictly adhering to OpenAI's API standard, Voice Mode enables powerful deployment flexibility:

• 🔀 Transparent Routing: Users can implement their own API proxies or gateways outside of Voice Mode to route requests to different providers based on custom logic (cost, latency, availability, etc.)
• 🎯 Model Selection: Deploy routing layers that select optimal models per request without modifying Voice Mode configuration
• 💰 Cost Optimization: Build intelligent routers that balance between expensive cloud APIs and free local models
• 🔧 No Lock-in: Switch providers by simply changing the BASE_URL - no code changes required

Example: Simply set OPENAI_BASE_URL to point to your custom router:

export OPENAI_BASE_URL="https://router.example.com/v1"
export OPENAI_API_KEY="your-key"
# Voice Mode now uses your router for all OpenAI API calls

The OpenAI SDK handles this automatically - no Voice Mode configuration needed!

┌─────────────────────┐     ┌──────────────────┐     ┌─────────────────────┐
│   Claude/LLM        │     │  LiveKit Server  │     │  Voice Frontend     │
│   (MCP Client)      │◄────►│  (Optional)     │◄───►│  (Optional)         │
└────���────────────────┘     └──────────────────┘     └─────────────────────┘
         │                            │
         │                            │
         ▼                            ▼
┌─────────────────────┐     ┌──────────────────┐
│  Voice MCP Server   │     │   Audio Services │
│  • converse         │     │  • OpenAI APIs   │
│  • listen_for_speech│◄───►│  • Local Whisper │
│  • check_room_status│     │  • Local TTS     │
│  • check_audio_devices    └──────────────────┘
└─────────────────────┘

• No microphone access: Check system permissions for terminal/application
• UV not found: Install with curl -LsSf https://astral.sh/uv/install.sh | sh
• OpenAI API error: Verify your OPENAI_API_KEY is set correctly
• No audio output: Check system audio settings and available devices

Enable detailed logging and audio file saving:

export VOICE_MCP_DEBUG=true

Debug audio files are saved to: ~/voice-mcp_recordings/

To save all audio files (both TTS output and STT input):

export VOICE_MCP_SAVE_AUDIO=true

Audio files are saved to: ~/voice-mcp_audio/ with timestamps in the filename.

• Website: getvoicemode.com
• GitHub: github.com/mbailey/voicemode
• PyPI: pypi.org/project/voice-mcp
• npm: npmjs.com/package/voicemode

MIT - A Failmode Project