The start-and-end mode is the one that is hard to get elsewhere: fixing both frames turns generation into an interpolation problem, which is far more controllable than a text prompt alone. Plan around the wait — a clip typically takes between 30 seconds and 5 minutes depending on complexity and load, so this belongs in a workflow that can tolerate a gap, not in a tight loop.
The official bridge to Vidu's video generation models. You describe the shot in the conversation, the server submits the job to the Vidu platform, and you get back a URL to the finished clip.
- Text-to-video from a written description alone
- Image-to-video, which animates a still you supply
- Reference-to-video, which carries a subject from your images into a new scene
- Start-and-end-to-video, which fixes both ends of the clip and generates the motion between them
- Generation parameters you can name in the request — model, style, duration, aspect ratio, resolution and movement amplitude
The PyPI package vidu-mcp, version 0.1.2, run through uvx. Python 3.10 or higher and uv installed. VIDU_API_KEY and VIDU_API_HOST go in the client config; the key comes from the Vidu platform. Generation is metered — API credits must be purchased separately before anything runs.
One command plus a key — uvx vidu-mcp, then supply credentials