Why WebMCP
An AI agent animates a 3D character on this page. You open your figure, type what it should do, and download the clip. The 20 tools sit inside the page where the figure runs, so the agent works on the same figure you are looking at.
Better for the user
No animation software, no rig setup. The page measures your figure itself — height, bone roles, sole points. The rest is one sentence in the chat, and you watch it happen.
What is new
Ask an agent for a backflip today and you get a figure that reaches through itself or barely moves. Not because the AI is too dumb. It has too little to work with: it does not know how tall the figure is, where its weight sits, how far a knee bends, or what its last move actually did.
This page gives it all of that. You type a sentence, the agent builds the motion, and what comes out is usable.
How it works
20 tools registered through document.modelContext.registerTool: describe the figure,
set poses and joint angles, check the result, render an image.
Everything the tools report is measured on the loaded model, not hardcoded. Answers carry numbers with units, and when the solver misses what was asked, it says by how much.
Built With
- javascript
- three.js
- webmcp
Log in or sign up for Devpost to join the conversation.