Everything runs locally against real hardware, so a recording is a file on your own disk rather than an upload. Worth knowing before you plan around it: text-to-speech is not implemented, so this is capture and playback, not speech.
A Python server that connects an MCP client to the audio hardware on your machine. It enumerates the input and output devices, records from a chosen microphone at a duration and sample rate you set, and plays audio back through the speakers.
- Every microphone and speaker on the system, listed with the index you address it by
- A recording from any input device, with duration and quality set per call
- The most recent recording played straight back
- An audio file on disk played through the speakers
- Text-to-speech is named in the README as a placeholder for later, not a working call
Python 3.8 or newer, and audio input and output devices on the machine. It runs as a local process, so the client and the hardware have to be on the same computer. A setup script generates the client configuration for you.
One command — npx -y @smithery/cli install @GongRzhe/Audio-MCP-Server --client claude