Claude PTT - Push-to-Talk Voice Input for Claude Code

A Claude Code plugin that enables voice input via push-to-talk. Hold a hotkey to record your voice, release to transcribe and insert text into Claude Code.

Features

• Push-to-talk: Hold Ctrl+Space to record, release to transcribe
• Dual transcription backends: OpenAI Whisper API and local whisper.cpp
• Cross-platform: Windows, macOS, Linux (X11 and Wayland)
• Automatic fallback: Falls back to alternative backend if preferred fails
• Visual feedback: Shows recording/transcribing status

Installation

From Marketplace

/plugin marketplace add aaddrick/claude-ptt
/plugin install ptt@claude-ptt-marketplace

Manual Installation

git clone https://github.com/aaddrick/claude-ptt.git
cd claude-ptt
npm install
npm run build

Configuration

Configuration is stored in ~/.claude/ptt-config.json:

{
  "hotkey": "Ctrl+Space",
  "whisper": {
    "openaiApiKey": null,
    "localModelPath": null,
    "preferredMode": "api",
    "enableFallback": true,
    "language": "en"
  },
  "audio": {
    "sampleRate": 16000,
    "silenceThreshold": 0.5
  },
  "keystroke": {
    "waylandBackend": "wtype"
  },
  "feedback": {
    "showRecordingIndicator": true
  }
}

Whisper Configuration

API Mode (Recommended for ease of use):

• Set OPENAI_API_KEY environment variable, or
• Set openaiApiKey in config

Local Mode (Recommended for privacy):

• Install whisper.cpp
• Set localModelPath to your model file

Setting Preferred Mode

Use preferredMode to choose which backend to try first:

• "api": Try OpenAI API first
• "local": Try local whisper.cpp first

Enable enableFallback to automatically try the other backend if the preferred one fails.

Platform Setup

Windows

No additional setup required. The plugin uses nut.js for keystroke simulation.

macOS

Grant accessibility permissions to your terminal application:

1. Open System Preferences > Security & Privacy > Privacy > Accessibility
2. Add your terminal app (Terminal.app, iTerm2, etc.)

Linux (X11)

Install libxtst for keystroke simulation:

sudo apt install libxtst-dev

Linux (Wayland)

Install one of the following for keystroke simulation:

# Option 1: wtype (recommended, no daemon required)
sudo apt install wtype

# Option 2: ydotool (requires daemon)
sudo apt install ydotool
sudo systemctl enable --now ydotool

# Option 3: dotool
# Build from source: https://sr.ht/~geb/dotool/

Audio Recording

The plugin uses system audio tools:

• Linux: arecord (ALSA, usually pre-installed)
• macOS/Windows: sox (install via brew install sox or download from http://sox.sourceforge.net/)

Usage

Starting the Daemon

# Via npm
npm start

# Or directly
node dist/daemon.js

Using with Claude Code

1. Start the daemon in a separate terminal
2. In Claude Code, hold Ctrl+Space to record
3. Speak your message
4. Release Ctrl+Space to transcribe
5. Text appears in your input for review
6. Press Enter to submit

MCP Tools

The plugin provides MCP tools for configuration:

• ptt_get_config: Get current configuration
• ptt_set_config: Update configuration
• ptt_get_status: Get daemon status
• ptt_get_platform_info: Get platform info and setup instructions

Troubleshooting

Hotkey not detected

• Linux: May need to run as root for global key capture
• macOS: Ensure accessibility permissions are granted
• All platforms: Check for conflicts with other applications

Keystroke simulation not working

• macOS: Check accessibility permissions
• Linux Wayland: Ensure wtype or ydotool is installed
• Linux X11: Ensure libxtst-dev is installed

Transcription fails

• API mode: Verify your OpenAI API key
• Local mode: Verify whisper.cpp is installed and model path is correct
• All modes: Check microphone permissions and audio recording

No audio recorded

• Check microphone permissions in system settings
• Verify audio recording tool is installed:
  • Linux: which arecord
  • macOS/Windows: which sox

Development

# Install dependencies
npm install

# Build TypeScript
npm run build

# Watch mode
npm run watch

# Run daemon
npm start

# Run MCP server
npm run mcp-server

License

MIT

Author

Aaddrick Williams aaddrick@gmail.com