generate-video takes the asynchronous path by default and switches to a blocking one when the request says to wait, while submit-video-task and get-video-task make the same thing explicit — which is what an agent loop that cannot sit on an open call actually needs. The image side is narrower by comparison: text, style, colour and ratio go in, and nothing here edits what comes back.
A four-tool server over Jimeng AI's text-to-image and text-to-video models, with the video side offered both as one blocking call and as a submit-then-collect pair.
- An image generated from text, with an illustration style, a colour and an aspect ratio as arguments
- A video generated from a prompt, returning asynchronously by default and synchronously when the request asks to wait for the result
- A video task submitted on its own, returning its task ID immediately
- The finished video collected later by task_id
JIMENG_ACCESS_KEY and JIMENG_SECRET_KEY — an access key pair for Jimeng AI, which is what every one of the four tools authenticates with.
One command plus a key — npm install -g jimeng-ai-mcp, then supply credentials