Add Mic support via VS Code Speech

Yes please! Dictation is a much nicer experience – especially for those with RSI.

Hey @spirefyio and @yoniholmes! 👋

Great news - Voice Mode provides the dictation capabilities you're looking for, and it works directly with Claude Code (no VS Code needed)!

Why Voice Mode is perfect for your needs:

✅ Native mic support - Works directly in Claude Code
✅ Fast dictation - Just speak naturally, no manual start/stop
✅ RSI-friendly - Reduce typing strain (great point @yoniholmes!)
✅ Smart editing - Say "Let's have a voice conversation" and iterate verbally

Quick setup:

# One-time installation
claude mcp add voice-mode --env OPENAI_API_KEY=your-key -- uvx voice-mode

# Start dictating
claude converse

Key advantages over VS Code Speech:

1. Integrated experience - No switching between apps
2. Two-way voice - Hear Claude's responses too
3. Natural pauses - Silence detection handles when you stop speaking
4. Cross-platform - Works on Linux, macOS, and Windows (WSL)

For RSI sufferers:

• Minimal keyboard use - just speak your prompts
• Voice-driven editing - describe changes verbally
• Configurable listening duration - take your time formulating thoughts

Voice Mode turns those detailed prompts into natural conversations. No more struggling with backspace or editing - just talk through your ideas!

Would love to help you both get started with hands-free coding!

So this is a free/open source option? I see the link to your GH. Very cool! I will def give it a try. Be curious how it works via MCP?

Yes @spirefyio , and it life changing for me (but only a few hundred people know about it so far).

Yes, despite the slick logo Voice Mode MCP is a partnership between one guy (OG) and Claude Opus 4.

After > 1 billion tokens and many late nights, this innovative approach has produced the worlds smartest Voice Assistant.

I converse with Claude and work on projects while walking the dog, driving, cleaning the house... I have my laptop in my backpack and my airpods in my ears. It's like "Blind Eye for the Dev Guy" or "You need to get out more!" - I am out more. :-)

There is no revenue model. It's totally free and open source. I've been working toward this for 18 months and using MCP was a surprising twist.

Please try it out: https://getvoicemode.com

Please share it: As an unfunded open source project I don't have budget to ask Fireship to do a video about it. Please Star, Like, Subscribe, tell your friends, etc.

Watch this - it's Voice Mode MCP loading into Claude Code and Gemini CLI (it should work anywhere MCP works)

https://youtu.be/y07nFEk9Q6M

<a href="https://youtu.be/y07nFEk9Q6M"><img width="3004" height="1857" alt="Image" src="https://github.com/user-attachments/assets/acf74df1-a44d-4e51-8846-b701408df9bd" /></a>

Looks very cool.. installed but I hope I don't need OpenAI API key as well to use it? It should just work with Claude Code right? I dont use OpenAI. Just Claude Code. I started it but not sure how to actually get it to listen. Doesn't seem to work as is.

To be up and running in minutes: Get an OpenAI Key from https://platform.openai.com
To run without OpenAI: Install your own ASR and TTS services

https://github.com/mbailey/voicemode?tab=readme-ov-file#-simple-requirements

<img width="1505" height="728" alt="Image" src="https://github.com/user-attachments/assets/076bf805-82d7-4b5f-aacd-4e5afe81c952" />

Ah.. that complicates it. Oh well.. looks like it could be useful. Be great if CC just added mic support somehow but probably cant.

Ah.. that complicates it. Oh well.. looks like it could be useful. Be great if CC just added mic support somehow but probably cant.

I reread your original post and see you were asking for an existing speech recognition tool built into VS code to be integrated with the Claude code plug-in? Dictation is great. For most people it makes them twice as fast as the fastest touch typer. That seems like a pretty easy thing to add that would just require users to push a button.

I don't know if you've watched any of the videos but voice mode MCP is a different base altogether. You literally don't need to look at the screen or touch the key board which means you can do voice coding while walking the dog or cooking dinner.

I tested out the "Push to Talk" functionality and also Voice Mode MCP and made this video demonstrating how much better Voice Mode MCP is. It provides a choice of voices, option to use local open source SST and TTS and much more.

I use Claude Code in the terminal and have not tried it in VS Code but suspect Voice Mode MCP would work the same as with GitHub Co-Pilot.

I made a screencast that demonstrates Voice Mode MCP in VS Code GitHub Copilot:

https://youtu.be/3sQRt2ITLOw?feature=shared&t=452

<a href="https://youtu.be/3sQRt2ITLOw?feature=shared&t=452"><img width="2111" height="1466" alt="Image" src="https://github.com/user-attachments/assets/e4596ebe-b1ac-4dd3-b1a6-206ac05b713e" /></a>

MCP is too slow. Try this instead: https://github.com/robdmac/talkito

@robdmac looks interesting. The video was very short though. Is it not using MCP/etc? How does it convert my voice audio into text? Is it using AI in the cloud (e.g. via API tokens/etc)? Or does the python code actually convert any talking to text in the app itself? I should add.. though I suspect there is no way to get a local app to convert audio to text easily.. I am a little leery of my audio chats going up to some server to convert to text, where my chats could be stored and perused/listened to by others or scanned by AI to figure out what I am doing. I am working on proprietary stuff with CC and don't really want my voice/chats stored and used in any way that can a) be used against me b) allow my "secret sauce" to be scanned/shared/stolen, etc. I am not entirely sure what happens to the audio on the MCP server it goes to (I looked at the source of your project and it appears you basically use MCP server behind the scenes??).

Thanks that's useful feedback and good to understand your concerns.

Sorry the video was done in a super rush to submit it for a hackathon. I'll update that soon.

It works by hijacking the input and output of Claude Code - which would be a lot easier if they open sourced it lol!

I do support MCP and I use that for a chrome extension mode so that it works with bolt.new/repel.it etc. However for Claude Code the MCP server is only used to configure TalkiTo so you can type "turn ASR off" or "change TTS to Amazon Polly" and it should know what to do.

It's essentially BYOK and chose your own provider. The ASR component that you are asking about is set by default to the free google on so yes that goes up to a cloud. There are options if you don't trust google (I recommend Assembly who give a lot of free credits). I'm actively working on integrating local model providers so for TTS ive recently added support for Kokoro and KittenTTS, both of which download the model to your device and run offline. I'll try and get a local ASR model like whisper added.

I share your privacy concerns and it's one of the reasons I did this as open source so people can verify that I'm not doing anything nefarious.. just sending it straight on to the providers.

Very cool. I love the ambition and look forward to the local run option. I am still relatively new with AI/models/llm/etc. Would I be able to use that with an LM Studio model and in server mode? If so would there be a specific model for voice to text to use? I can also run llama.cpp if that is preferred.

For the coding model, I'm planning on supporting for other agents coders such as gemini cli or aider and some of which (such as aider) will support local models. I'm just focussing on getting the Claude Code (which will presumably always be limited to Claude's online models) integration perfect first because for me thats the one thats ahead of the game and I use all the time.

As for the voice to text and text to voice, TalkiTo already supports most of the major providers there so you can mix and match. Right now you could, for example, pick Assembly ASR with Mac OS system TTS with Claude Code. Its currently most TTS providers supported with most ASR providers supported with Claude Code and a few browser based ones supported.

Small update @spirefyio I added support for Whisper so now (along with Kokoro TTS) you can essentially have the full TTS and ASR features without calling any service other than Claude's: talkito --tts-provider kokoro --asr-provider local_whisper claude

https://github.com/robdmac/talkito

I just wanted to address the claim that "MCP is too slow for voice". I've not seen anyone provide evidence to support that claim.

MCP is not the bottleneck. Sub-second TTS and STT is possible. (Metrics included in single take video)

https://www.youtube.com/watch?v=A8QP42iCeyY

<a href="https://www.youtube.com/watch?v=A8QP42iCeyY"><img width="1429" height="795" alt="Image" src="https://github.com/user-attachments/assets/50c4074e-04b5-4df5-9137-201257e7baef" /></a>

You can get faster performance by using less capable models or more powerful local compute. I recorded this on an M4 Mac Mini with Opus and Whisper Large-v2 for high accuracy.

VoiceMode MCP now has CLI tools for managing local voice services (Kokoro and Whisper.cpp).

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

Is anyone working on this?

This issue has been automatically closed due to 60 days of inactivity. If you're still experiencing this issue, please open a new issue with updated information.

This issue was closed incorrectly despite recent human comments. This behavior of the bot is reported at https://github.com/anthropics/claude-code/issues/16497. Please upvote that issue, so maybe it gets noticed.

This would be nice to have

+1, a native speech-to-text would be a great addition to claude code

VoiceMode is the leading open source solution for natural voice conversations with Claude Code:

• 700 GitHub ⭐
• 132K downloads

<a href="https://github.com/mbailey/voicemode"><img width="1280" height="720" alt="Image" src="https://github.com/user-attachments/assets/8a9de401-edee-47a3-acf3-b9cfe1945ddb" /></a>

Quickstart

# Add the plugin marketplace
claude plugin marketplace add mbailey/plugins

# Install VoiceMode plugin
claude plugin install voicemode@mbailey

## Install dependencies (CLI, Local Voice Services)

/voicemode:install

# Start talking!
/voicemode:converse

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.