It comes from Daisys rather than a third party, and the credential model is straightforward: account email and password in the environment, output path under your control. The install failure people actually hit is PortAudio — the README carries a separate troubleshooting note for Linux because of it, so do that step before anything else.
The Daisys team's own MCP server for the Daisys platform. It signs in with your Daisys account, produces audio, and stores the resulting files locally at a path you set. Playback goes through PortAudio, which is why the install has a system-level step on macOS and Linux before the Python package matters.
Audio produced through your own Daisys account, written to the directory named by `DAISYS_BASE_STORAGE_PATH` so the output stays on your machine and is addressable afterwards. The project describes itself as beta with no stable release yet, and its full integration test suite is marked as costing tokens on the Daisys platform when run — a signal that generation is metered against your account.
A Daisys account with a username and password, supplied as `DAISYS_EMAIL` and `DAISYS_PASSWORD`, plus `DAISYS_BASE_STORAGE_PATH` for where audio files land. PortAudio must be present first: `brew install portaudio` on macOS, `sudo apt install portaudio19-dev libjack-dev` on Linux. Then run `uvx daisys-mcp` over stdio, or build from source with `uv venv` and `uv pip install -e .` and launch `daisys_mcp.server`. `uv run mcp dev daisys_mcp/server.py` opens the Inspector for debugging.
One command plus a key — uvx daisys-mcp, then supply credentials