The server only sees the directory you bind in, which makes the blast radius of handing an assistant your microphone recordings a decision you make in the config rather than a matter of trust. The one thing to be deliberate about is cost: starting a job consumes your mocoVoice allowance, so the file listing and format check are worth running before the job, not after a failed one.
An MCP server that connects a client to mocoVoice for transcription. It works over a directory you mount, so the assistant can see which media files are available, start a job on one, and fetch the transcript when it is done.
- A list of the audio and video files it can actually reach, limited to the directory you mounted — `SHOW_AVAILABLE_FILES`
- The audio and video formats it accepts, before you convert anything — `SHOW_AVAILABLE_FORMATS`
- Transcription job submission against a file, given as an absolute path — `START_TRANSCRIPTION_JOB`
- Result retrieval by transcription id, and a list of every past transcription with its status — `SHOW_TRANSCRIPTION_RESULT`, `CHECK_TRANSCRIPTION_STATUSES`
- A capability listing of the server itself — `SHOW_USAGE`
- Uploads of up to 5GB, covering .wav, .mp3, .m4a, .caf, .aiff, .wma, .flac, .ogg and .aac for audio and .avi, .mp4, .rmvb, .flv, .mov and .wm for video
A mocoVoice API key with READ and WRITE permissions, as `MOCOVOICE_API_KEY`, plus `MOCOVOICE_API_URL` pointing at https://api.mocomoco.ai/api/v1. Docker, since the documented setup runs `ghcr.io/mocomoco-inc/mocovoice-mcp-server` with your media directory bind-mounted to /workspace — files are addressed by absolute path. Transcription is billed as mocoVoice API usage, and starting a job consumes quota. The package is `moco-voice-mcp` (0.1.0 in pyproject); it is offered as a beta.