mcp_servers

This MCP server provides a voice recording and transcription workflow using OpenAI's Whisper model. It can run as a standalone MCP server or be used as part of Goose AI workflows via the Goose extension. The server records audio from the default microphone, transcribes it with Whisper, and exposes tools to start recording, stop and transcribe, or record for a specified duration. You can adjust the Whisper model (tiny.en, base.en, small.en, medium.en, large) and settings such as sample rate and maximum duration. The server also supports integration with Goose by configuring an explicit command path in Goose Settings and using the provided prompts to incorporate transcription results into conversations. Tools available include start_recording, stop_and_transcribe, and record_and_transcribe for flexible recording and transcription flows.

Prerequisites:\n- Python 3.12 or higher\n- Git\n- pip (Python package manager)\n\nInstallation steps:\n1) Clone the repository from GitHub:\n git clone https://github.com/DefiBax/voice-recorder-mcp.git\n2) Navigate into the project directory:\n cd voice-recorder-mcp\n3) Install in editable mode (from source):\n pip install -e .\n4) Ensure the executable is on your PATH (the installation provides a CLI named voice-recorder-mcp). You can verify by running:\n voice-recorder-mcp --help\n\nOptional:\n- Create a virtual environment before installation if you prefer isolation:\n python -m venv venv\n source venv/bin/activate # on Unix/macOS\n venv\s reactivate # on Windows\n- For Goose integration, follow the Goose setup steps outlined in the README to point Goose at the voice-recorder-mcp executable.\n

Environment variables can customize behavior without altering command line arguments:\n- WHISPER_MODEL: choose the Whisper model (e.g., small.en, medium.en, base.en, tiny.en, large).\n- SAMPLE_RATE: set the audio sample rate (default 16000 in some configurations; use 44100 for higher fidelity).\n- MAX_DURATION: cap recording duration in seconds (e.g., 120).\nCommon issues include microphone permissions, initial model download requiring internet access, and ensuring the path to the executable is correct in Goose. If audio quality is poor, adjust the sample rate and verify the microphone is accessible by your OS. The MCP Inspector tool can be used to test the server command and response structure before full integration.\n