VoiceMode

VoiceMode turns Claude into a voice assistant you can actually talk to. It wires up OpenAI's speech-to-text and text-to-speech APIs, plus LiveKit for real-time audio streaming, so you can have proper spoken conversations with Claude through your microphone. The setup is straightforward if you're already running Claude Desktop, and it works with Claude Code too. I'd reach for this when I'm debugging something complex and need to think out loud, or when I'm away from the keyboard but still want to work through a problem. The hands-free coding angle is real: you can describe what you want, hear Claude's response, and iterate without typing. It's also genuinely useful for accessibility workflows, though the reliance on OpenAI's APIs means you're paying per minute of audio processed. The LiveKit integration is what makes this more than just a dictation tool. You get proper back-and-forth conversation, not just one-shot commands. That said, you'll need an OpenAI API key with credits, and you'll burn through them faster than you might expect if you're chatty. The server handles local mic input cleanly, but you're still dependent on network latency for the speech processing. Quirks: it's a community project, so expect some rough edges. The documentation assumes you know your way around MCP server configs. If you've never set up an MCP server before, this isn't the easiest first one to try. But if you're already comfortable with the ecosystem, it's a solid addition. Who shouldn't bother: anyone hoping for offline voice processing, or anyone who finds the idea of talking to their computer awkward. This is for people who already think out loud when they code, or who genuinely need hands-free access. If you're happy typing, stick with the keyboard.