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.