Back to Blog
Developer Productivity March 26, 2026 5 min read

Figma Opens Its Canvas to AI Agents — Claude Code, Cursor, and Copilot Can Now Design

Figma's new use_figma MCP tool lets AI coding agents write directly to the design canvas, creating and editing real components from your design system. Open beta started March 24, free for paid plan users during the beta period.

Figma Opens Its Canvas to AI Agents — Claude Code, Cursor, and Copilot Can Now Design

Figma’s canvas is now writable by AI agents. The use_figma MCP tool, which entered open beta on March 24, lets Claude Code, Cursor, GitHub Copilot, and a dozen other MCP-compatible clients create, edit, delete, and inspect any element in a Figma file — pages, frames, components, variants, variables, styles, text, and images.

The distinction from the existing generate_figma_design tool matters. That tool converts live HTML or websites into editable Figma layers — useful for syncing code back to design when they drift apart. The use_figma tool works the other direction: agents operate inside Figma natively, with access to your design system’s existing components and variables. When an agent needs a button, it reaches for the button in your component library. It doesn’t generate a generic one from scratch.

That shift — from output to system-aware authorship — is the whole point. The quality of what an agent produces is now bounded by the quality of your design system. A mature, well-organized system yields useful output. A disorganized one yields garbage. Figma is effectively making design system hygiene a competitive advantage in the AI era.

Skills are the accompanying abstraction. They’re markdown files that tell agents how to behave in Figma: which tools to call, how to sequence them, what conventions the team follows. Nine community-built skills shipped alongside the open beta, covering tasks like building component libraries, generating new screens, producing accessibility specs, and bridging design tokens to code. Skills can also define self-healing loops — where the agent screenshots its output, compares it to the target, and iterates on mismatches.

Skills don’t require plugin development. They’re just markdown. Any designer or developer can write one, share it, or fork an existing one from figma.com/community/skills.

Supported clients at launch include Augment, Claude Code, Codex, GitHub Copilot CLI, Copilot in VS Code, Cursor, Factory, Firebender, and Warp. The MCP server runs natively in Figma, meaning it inherits Figma’s security model and gets access to surfaces like Code Connect, Figma Draw, and FigJam through the Plugin API.

Pricing is straightforward: it’s a paid API, free during beta. Access requires a Full or Dev seat on a paid Figma plan. Dev seats are read-only outside of Drafts.

The roadmap items Figma called out: image support, custom fonts, and progress toward full Plugin API parity. The longer-term implication is a design workflow where agents handle the mechanical layer — translating requirements into Figma-native structure — while designers focus on the decisions that require judgment. Whether that plays out as a productivity multiplier or as a reduction in design headcount depends on how teams choose to use it. The capability is real either way.

Figma MCP AI Agents Design Tools