voice-mode

What it does

VoiceMode adds natural voice conversations to Claude Code through an MCP interface. You speak, Claude responds aloud — enabling hands-free interaction when typing or looking at a screen isn't practical. The server supports both cloud speech services (OpenAI API) and fully local alternatives (Whisper.cpp for speech-to-text, Kokoro for text-to-speech), letting you trade latency for privacy as needed.

Who it's for

Developers and Claude Code users who want to interact while their hands or eyes are unavailable: debugging while cooking, debugging while walking, giving your eyes a break after hours of coding, or holding a coffee without setting it down. Works on Linux, macOS, Windows (WSL), and NixOS.

Common use cases

• Debug or refactor code while cooking or preparing food
• Ask Claude for code review or guidance while walking to a meeting
• Get explanations or brainstorm solutions without looking at a screen
• Run conversations entirely offline using local Whisper and Kokoro services
• Test ideas or pseudocode by speaking them aloud and hearing feedback

Setup pitfalls

• Secrets found in package during scanning — use environment variables for OPENAI_API_KEY and other credentials, never hardcode them.
• Platform-specific system dependencies: macOS requires ffmpeg and portaudio (via Homebrew); Linux (including WSL) requires ffmpeg, portaudio19-dev, and WSL specifically needs pulseaudio packages for microphone input.
• Microphone and speaker permissions must be granted at the OS level; some systems require explicit terminal/app permissions before audio input works.
• High risk classification — reads and writes audio files to disk (when VOICEMODE_SAVE_AUDIO=true) and makes network calls for remote speech services; use caution with untrusted code paths.