---

name: voice-interaction description: Enable voice conversations with Claude Code - speak commands and hear responses tools:

• mcp__voice__voice_listen
• mcp__voice__voice_speak
• mcp__voice__voice_conversation
• mcp__voice__voice_status

---

Voice Interaction Mode

Speak to Claude Code and hear responses. Perfect for hands-free coding, accessibility, or when you're away from the keyboard.

Quick Start

1. Check voice is working:

   CopyUse voice_status tool
   

2. Listen to user:

   CopyUse voice_listen with duration=5
   

3. Speak response:

   CopyUse voice_speak with text="Hello, I heard you!"
   

Voice Conversation Flow

When in voice mode, follow this pattern:

Copy1. Speak a prompt/question to user
2. Listen for their response
3. Process what they said
4. Speak your response
5. Repeat

Example Conversation

CopyClaude: voice_speak("What would you like me to help with?")
Claude: voice_listen(duration=10)
→ User said: "Can you find all the TODO comments in my code?"
Claude: [runs grep for TODOs]
Claude: voice_speak("I found 5 TODO comments. The first one is in main.py line 42 about fixing the database connection.")

Commands via Voice

Users can say things like:

• "Search for [pattern] in [file/directory]"
• "Open [filename]"
• "Run the tests"
• "Fix the error in [file]"
• "What does [function name] do?"
• "Commit these changes with message [message]"
• "Stop" or "Cancel" or "Never mind"

Voice Response Guidelines

When speaking responses:

DO

• Keep responses concise (< 30 seconds of speech)
• Summarize long outputs ("Found 15 files, the most relevant are...")
• Confirm actions before running them
• Ask clarifying questions for ambiguous requests
• Spell out unusual names/symbols

DON'T

• Read entire file contents aloud
• Speak raw error messages (summarize them)
• Give long explanations (offer to show on screen instead)
• Assume - ask if unclear

Handling Voice Commands

Parse user speech and map to actions:

Error Handling

If transcription is unclear:

Copyvoice_speak("I didn't catch that. Could you repeat?")
voice_listen(duration=10)

If command is ambiguous:

Copyvoice_speak("Did you mean X or Y?")
voice_listen(duration=5)

Voice Settings

Control voice behavior via environment variables:

Copy# Whisper model size (tiny/base/small/medium/large)
export WHISPER_MODEL=base

# Default recording duration
export VOICE_DURATION=5

Accessibility Notes

Voice mode is designed for:

• Hands-free operation
• Screen reader compatibility
• Reduced visual dependency

When voice is active, also output text for screen readers.

Troubleshooting

"No audio available"

Copypip install sounddevice numpy

"Whisper not found"

Copypip install openai-whisper

"No TTS on Linux"

Copypip install pyttsx3

Poor transcription

• Speak clearly, closer to microphone
• Reduce background noise
• Try longer duration
• Use larger Whisper model: export WHISPER_MODEL=small