You can enable natural voice interaction with Claude Code through an MCP server that runs locally to capture your voice commands and send them to Claude Code for real-time responses. This setup lets you wake Claude with a spoken phrase, dictate commands, and see rapid code-related results while keeping privacy in mind by activating listening only when prompted.

How to use

To use this MCP server, start the local voice command server and connect it with your MCP client. You will be able to wake Claude with a wake word, speak commands like creating or modifying code, and let Claude respond with code updates or explanations. Use the MCP client to trigger voice mode, issue commands, and render Claude Code’s responses in your editor or IDE.

Practical usage patterns you can try after setup: say the wake phrase to begin listening, then issue a command such as asking Claude to generate a Python function or to add error handling. You can queue multiple commands in sequence and Claude will respond to each in order. When you are done, tell Claude to stop listening so it ceases continuous capture.

Example interactions you can reproduce with your MCP client. You start voice mode, request a function, request enhancements, and then stop voice mode. The server handles the sequence and your editor updates accordingly.

How to install

Prerequisites you need before installing the MCP server are: Python 3.8+ and a working microphone on macOS, Windows, or Linux, plus the Claude Desktop app.

# 1. Clone the project
git clone https://github.com/JWitcoff/Claude_Chat.git
cd Claude_Chat

# 2. Create a virtual environment
python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# 3. Install dependencies
pip install -r requirements.txt

# 4. Test your microphone
python tests/test_microphone.py

# 5. Configure Claude Desktop (Sprint 2+) with MCP
# Add to Claude Desktop config (path shown exactly as an example)
# The following is a JSON snippet you will place in the config file

""" { "mcpServers": { "voice-commands": { "command": "python", "args": ["-m", "mcp_server.voice_server"], "cwd": "/path/to/Claude_Chat", "env": { "PYTHONPATH": "/path/to/Claude_Chat" } } } } """

6. Restart Claude Desktop to apply the MCP server configuration and enable voice commands.

## Configuration

Create a .env file to tune recognition behavior and PR features. The following settings reflect the available options you can use.

Recognition Settings

VOICE_BACKEND=google # Primary: google, whisper, openai VOICE_ENERGY_THRESHOLD=4000 # Microphone sensitivity WAKE_WORDS=hey claude,claude

Optional API Keys (for premium features)

OPENAI_API_KEY=sk-... # For ultra-low latency `

## Troubleshooting

If you encounter issues with your microphone or recognition, try these steps. List available microphones, calibrate ambient noise, and adjust the energy threshold as needed.

List available microphones

python -c "import speech_recognition as sr; print(sr.Microphone.list_microphone_names())"

Calibrate for ambient noise

python tests/calibrate_mic.py

Notes

This MCP server uses a Python-based backend to listen for voice commands and forward them to Claude Code via the MCP protocol. It is privacy-focused in that listening occurs only after the wake word is detected. You can switch backends or adjust sensitivity to suit your environment.

Available tools

google_speech_recognition

Uses Google Speech Recognition to transcribe spoken input with low latency when selected as the primary backend.

whisper_backend

Uses OpenAI Whisper as an alternative transcription backend for offline or privacy-preserving scenarios.

openai_realtime

Provides realtime transcription and command processing using OpenAI services for ultra-low latency.

mic_calibration

Calibrates the microphone for ambient noise and determines an appropriate energy threshold.