Voice mode

Yes, however I highly recommend to make it configurable from day 1, including ability to provide custom STT voice command,

as person living in country of speaking multiple languages and accents I find ability to low level configure (like in Aider) STT critical!

Also for privacy reasons some users may prefer to do STT locally.

I had an idea to implement MCP with text-to-speech capabilities, together with OS native dictation setup like this should be pretty close to the Voice, mode:)

Have you tried using your laptop's built in voice mode? What do you envision would be the difference from having this built-in? I know some of the Claude Code team uses Mac Dictation for this

Have you tried using your laptop's built in voice mode? What do you envision would be the difference from having this built-in? I know some of the Claude Code team uses Mac Dictation for this

Not sure about those guys, but I'm using WSL2 Ubuntu on Windows 11, personally.

For Linux users https://github.com/KoljaB/RealtimeSTT , https://github.com/ggml-org/whisper.cpp , and many more... (depending if just cpu or also gpu is available, those can also work on just CPU)

Hey @nullbio! 👋

Great timing on this issue! I'm the developer of Voice Mode, an MCP server that already provides exactly what you're looking for - natural voice conversations with Claude Code.

What Voice Mode offers today:

✅ Voice conversations - Speak to Claude and hear responses back
✅ Configurable STT/TTS - Works with OpenAI, local Whisper.cpp, and other OpenAI-compatible services
✅ Smart silence detection - Automatically stops recording when you pause (no manual stop needed!)
✅ Multiple transports - Local microphone or LiveKit for room-based communication
✅ Emotional TTS - Using OpenAI's gpt-4o-mini-tts model for expressive speech (as @nullbio suggested!)

Quick setup:

# Install with Claude Code
claude mcp add voice-mode --env OPENAI_API_KEY=your-key -- uvx voice-mode

# Start talking
claude converse

Then just tell Claude: "Let's have a voice conversation" and you're talking! 🎤

Recent improvements:

We just added min_listen_duration parameter to prevent premature cutoffs - perfect for when you need thinking time before speaking. The system adapts to your speaking patterns.

Local/private options:

As @gwpl suggested, we support fully local processing:

• Whisper.cpp for local STT (no cloud needed)
• Kokoro for local TTS with multiple voices
• All with the same OpenAI-compatible API

Check out our demo video to see it in action! (Fun fact: it's unscripted and Claude helped edit it using whisper.cpp transcripts + ffmpeg!)

Would love your feedback on what features would make this even better for reducing typing fatigue. The emotional tone detection is already partially there with our TTS instructions feature - what other aspects would help?

Hey @nullbio! 👋 Great timing on this issue! I'm the developer of Voice Mode, an MCP server that already provides exactly what you're looking for - natural voice conversations with Claude Code. ## What Voice Mode offers today: ✅ Voice conversations - Speak to Claude and hear responses back ✅ Configurable STT/TTS - Works with OpenAI, local Whisper.cpp, and other OpenAI-compatible services ✅ Smart silence detection - Automatically stops recording when you pause (no manual stop needed!) ✅ Multiple transports - Local microphone or LiveKit for room-based communication ✅ Emotional TTS - Using OpenAI's gpt-4o-mini-tts model for expressive speech (as @nullbio suggested!) ## Quick setup: # Install with Claude Code claude mcp add voice-mode --env OPENAI_API_KEY=your-key -- uvx voice-mode # Start talking claude converse Then just tell Claude: "Let's have a voice conversation" and you're talking! 🎤 ## Recent improvements: We just added min_listen_duration parameter to prevent premature cutoffs - perfect for when you need thinking time before speaking. The system adapts to your speaking patterns. ## Local/private options: As @gwpl suggested, we support fully local processing: Whisper.cpp for local STT (no cloud needed) Kokoro for local TTS with multiple voices * All with the same OpenAI-compatible API Check out our demo video to see it in action! (Fun fact: it's unscripted and Claude helped edit it using whisper.cpp transcripts + ffmpeg!) Would love your feedback on what features would make this even better for reducing typing fatigue. The emotional tone detection is already partially there with our TTS instructions feature - what other aspects would help?

Great job mate! Thank you for making this. Going to give this a try tomorrow.

I'll also throw in another MCP/Wrapper solution: https://github.com/robdmac/talkito

Just pip install, talkito claude then say "start voice mode" (the Slack and WhatsApp etc is all optional)

Demo of Voice Mode MCP: Claude and Gemini talking discussing Gemini CLI

https://youtu.be/BfFshd7v4rw?si=T2gHTTZ4DI_DyKkq

Extended multilingual support a few hours ago.

I agree that voice mode would be amazing. However, we don't plan to implement a voice mode in Claude Code's application code unless Claude itself supports audio in the future. Instead, we plan to offer extension points (MCP, hooks) for other providers to provide this functionality.

@dicksontsai whilst I should be glad that TalkiTo won't be made redundant by an update, I have to point out that MCP based solutions are not going to cut it (no offense Mike). A voice interface is like a UI interface and needs to be super responsive. You wouldn't make the GUI elements go through an MCP tool call. This is why for TalkiTo I had to wrap Claude Code and intercept the input and output (although MCP is fine for configuring the interface). If you open sourced Claude Code I could make this voice interface much smoother...

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.