Figma opens direct AI agent design on canvas, introduces skills for guided workflows

Figma now lets AI coding agents design directly inside Figma files through its MCP server, so they can work on the canvas itself instead of generating code outside the app. Through the use_figma tool, compatible agents such as Claude Code and Codex can create and update design assets while using existing context like design systems, components, variables, layout structure, and team conventions.

Figma also introduced skills, which are markdown based instructions that shape how agents operate in Figma by defining steps, conventions, and workflow logic. They can be created or edited without code. At launch, Figma shared nine example skills focused on tasks like building component libraries, generating new designs, producing accessibility specs, coordinating workflows, and connecting code with design tokens. Skills can also support self healing loops, where agents compare, refine, and update designs based on real structure instead of only flat visuals.

Figma says it plans to expand this with native AI on canvas, better skill sharing, and support for images and custom fonts, which are not available yet. The feature currently works with several MCP clients, is free during beta, and is expected to become a paid API later. According to Figma, the MCP native setup also carries over existing security and reliability standards.