The appeal is breadth without integration work: image, video, music and 3D across a dozen named models, all behind a single PiAPI key. The contract-backed tools are the interesting engineering — a tool exists for every committed model-and-task pairing, and a drift detector keeps that surface aligned with PiAPI's catalog as it changes. The practical caveat is timeouts: video and other heavy jobs can outrun a client's default limit, so budget for that when wiring it up.
A TypeScript MCP server that fronts PiAPI's media-generation API, so an assistant can create images, video, music and 3D models directly. It ships hand-written convenience tools plus a contract-backed tool for every committed PiAPI model-and-task pairing, kept in sync with PiAPI's versioned catalog by a drift detector.
- Image generation from text or image prompts with Flux, and Midjourney imagine
- Video generation from text or image prompts — Hunyuan, Skyreels, Wan, Kling and Luma Dream Machine
- Music generation with Suno, and MMAudio music from video, plus zero-shot TTS voice generation
- 3D model generation from an image with Trellis
- A contract-backed tool per committed PiAPI capability, named `piapi_<model>_<task_type>`, generated from the committed catalog
- A drift detector that diffs the catalog against PiAPI Manager's versioned contract to flag new, deprecated or changed APIs
A PiAPI API key in `PIAPI_API_KEY`, from the PiAPI workspace. Node.js 16.x or higher. Install via Smithery, or clone plus `npm install` and `npm run build`, which produces `dist/index.js` for the client to launch with node. The package.json names it `piapi-mcp-server` at 1.0.0. The README warns that time-consuming jobs like video generation may not finish within a client's timeout — the MCP Inspector, for instance, defaults to 10 seconds, which it suggests raising for generation tasks.
One command plus a key — npx -y @smithery/cli install piapi-mcp-server --client claude, then supply credentials