Show HN: Local speech-to-text MCP server for Claude on Linux

#1

Post by marcindulak » Thu, Sep 11, 2025, 7:28 AM UTC

Link: https://github.com/marcindulak/stt-mcp-server-linux

A push-to-talk speech transcription setup for Linux, using a 500-line Python MCP server. Runs locally in Docker, no external API calls. Press Right Ctrl to record audio; Whisper transcribes it, and the transcript is sent into Claude running in a Tmux session.

Show HN: Local speech-to-text MCP server for Claude on Linux
github.com

Re: Show HN: Local speech-to-text MCP server for Claude on Linux

#2

Post by marcindulak » Thu, Sep 11, 2025, 9:15 AM UTC

I'm a Linux user and wanted to have a speech-to-text functionality in Claude, so I can talk to it, like Armin Ronacher https://www.youtube.com/watch?v=bpWPEhO7RqE#t=5m37s demonstrates on macOS. I was not able to find a small codebase doing this, that I can understand.

The project I'm submitting is about 500 lines of Python, and is packaged as Docker, so facilitate the setup. When creating the project I added some security measures, like running the Docker container as non-root, and performing Whisper output sanitization before passing it to Claude.

Thes setup is Linux-only due to `/dev` device dependencies.

Re: Show HN: Local speech-to-text MCP server for Claude on Linux

#3

Post by marcindulak » Sat, Sep 13, 2025, 5:10 AM UTC

FYI: discussed, and added to the curated list of Claude tools at https://github.com/hesreallyhim/awesome-claude-code/issues/1...