The useful part of the documentation is the constraint table: length and resolution are not free choices, they depend on which model you pick, and getting that wrong is the usual first failure. Generation is asynchronous by nature, so plan on a poll rather than a single blocking call.
An MCP server for Vidu's image-to-video API. Generation is asynchronous, so the surface is three steps: upload an image if you have a local file, start a generation from an image URL with a prompt and model settings, and poll the returned task id until the clip is ready.
- Image-to-video generation from an `image_url`, with an optional text `prompt` of up to 1500 characters, a `seed` for reproducibility, `movement_amplitude` to control how much motion is in frame, and an optional `callback_url` for asynchronous notification
- Model choice between viduq1, vidu1.5 and vidu2.0, with the length and resolution constraints each one imposes — viduq1 does 5 second clips at 1080p only, while vidu1.5 and vidu2.0 do 4 or 8 seconds, with 8 second output limited to 720p
- Background music on 4 second clips through the `bgm` flag
- A status check that takes the `task_id` returned by generation and reports progress along with credit usage
- An upload step taking `image_path` and `image_type`, for png, webp, jpeg or jpg files
A Vidu API key from vidu.com, supplied as `VIDU_API_KEY`. Node.js v14 or newer; the package is `vidu-mcp-server` (1.0.0 in package.json). Build with `npm run build` and point the client at `build/index.js`, or install through Smithery. Uploads are capped at 10MB, while images referenced by URL may be up to 50MB.
One command plus a key — npx -y @smithery/cli install @el-el-san/vidu-mcp-server --client claude, then supply credentials