Figma launches on‑canvas AI agents
Figma opened an open beta for AI agents that can design directly on the canvas using the new use_figma MCP tool—positioning agents as part of production design workflows. The demo shows agents creating and editing components in context. (x.com)
Figma’s official blog post announcing the MCP server was authored by Product Director Matt Colyer and published March 24, 2026, and it says the capability will move to a usage‑based paid model after the beta period. The new use_figma MCP tool gives selected MCP clients write access so agents can generate and modify design assets that are linked to a team’s design system, and Figma named Anthropic’s Claude Code and OpenAI’s Codex as early MCP clients in the announcement. Figma describes “skills” as markdown files that package team conventions, sequencing, and rules so agents follow established design system practices when they create or edit components. The platform keeps the generate_figma_design tool alongside use_figma so agents can import HTML or live UI into editable Figma layers and then have use_figma edit those frames using the project’s components and variables. The Figma MCP Server Guide on GitHub documents implementation details, shows the repo with community plugins and a “skills” folder, and lists rate limits and plan differences — for example, Starter/View/Collab seats are limited to up to six tool calls per month while some write tools are exempt from rate limits. Figma’s help center lists supported MCP clients and setup instructions (Claude Code, Codex, Cursor, VS Code), points to a Figma Community library of reusable skills, and publishes step‑by‑step remote and local MCP server setup guidance for teams.
