Cloning, stem separation, transcription and TTS normally mean four services and a GPU. Here they are ten tools on one endpoint, and the heavy ones return a job id rather than holding the conversation open.
A hosted audio-AI service reached over a single Streamable-HTTP endpoint at https://mcp.audiopod.ai. Nothing runs on your machine, and the long jobs hand back a job id you poll instead of blocking.
- Speech in dozens of languages, from a large library of stock voices or from a voice you cloned
- A voice cloned from a reference clip of a few seconds, and existing recordings converted to it
- Music from a text prompt: songs, instrumentals, rap or vocal stems
- A track split into stems — vocals, drums, bass — or two-stem mode for karaoke
- Each speaker in a recording isolated onto a separate track
- Transcription with word-level timestamps and speaker diarization
- Background noise removed with the voice character preserved
- Format conversion across mp3, wav, flac, ogg, m4a, mp4 and mov
- Status polling for the long-running jobs, which is free
An AudioPod API key sent as the X-API-Key header; OAuth2 and JWT also work. One HTTP entry in your client config — no install, no local models, no GPU. Free keys are issued from the AudioPod dashboard.
One command plus a key — claude mcp add --transport http audiopod https://mcp.audiopod.ai \ --header "X-API-Key: ap_YOUR_KEY" --scope user, then supply credentials