AI coding tools are good at code, but animation is hard to describe in code alone. Without access to the design, an agent guesses timing and states. MCP, the Model Context Protocol, fixes this by giving the agent direct access to a tool through a standard interface.
An MCP server is a small service that exposes actions an agent can call. For animation, that might mean creating a component, adding a state, changing a spring or reading which inputs a file has. The agent decides when to call these actions based on your prompt.
Setup is usually short. You add the server to your client's MCP configuration, for example with Claude Code's mcp command or in Cursor's MCP settings, then authenticate if needed. After that you can ask in plain language: "make a toggle with a spring and a pressed state" or "slow down the success animation".
ExodeUI, which we make, has an MCP server for this. Agents can create and edit animations with states, spring physics, inputs and events, and the result is an open .exode file you can open in the editor and play in your app. Design tools such as Figma and Penpot have their own MCP servers for layout and design context.
Treat agent output as a draft. Check the motion in the editor, especially timing, reduced-motion behaviour and every state, before it ships.