marcindulak/stt-mcp-server-linux

Implements push-to-talk speech-to-text via OpenAI Whisper (tiny model by default) with keyboard monitoring through `/dev/input` and audio capture via `/dev/snd`, running in a Docker container. Supports both MCP protocol integration with Claude Code and standalone Tmux mode, injecting transcribed text directly into terminal sessions using Tmux send-keys. Architecture uses modular composition with pluggable transcription engines (Whisper/Vosk) and output handlers, coordinated by an async event loop that keeps the server responsive during background transcription.