Adds three microphone tools to your MCP client: list available audio devices, capture audio to WAV files, and run a full offline voice pipeline. The voice_query tool chains capture, whisper.cpp transcription, and local Ollama inference so you can speak questions and get LLM responses without any cloud services. Built on decibri for cross-platform audio capture with no external dependencies like ffmpeg. You specify recording duration upfront since there's no voice activity detection yet. Works with Claude Desktop, Cursor, and other MCP clients. Useful when you want voice input without API calls or when building offline voice workflows.
claude mcp add --transport stdio decibri-mcp-listen uvx mcp-listen