Figma launched an open beta of the new MCP tool use_figma - now AI agents can draw designs directly on the canvas in Figma, rather than just reading from it.
Previously, the Figma MCP server worked one way - the agent could read the design and turn it into code. Now it's a two-way process: the agent can create and modify UI elements tied to your design system, and then you can refine the result by hand in Figma.
It works with Claude Code, Codex, Copilot, Cursor, Warp, and other MCP clients. During the beta period, use_figma is available for free - it will become a paid API later.
Interestingly, Figma added the concept of skills - essentially instructions that teach the agent to work on specific design tasks. You can share them and create your own.