Most AI coding agents can describe a UI or generate front-end code, yet they stay blind to the actual Figma file the design lives in. This project closes that loop: it hands an agent a live, two-way channel into Figma, so it reads the real layer tree and writes changes straight back instead of working from a static screenshot or export.
What Sets It Apart
- A live bridge, not a one-shot export. An MCP server, a WebSocket relay, and a Figma plugin form a round trip, so the agent operates on the running document — current selection, real node IDs, actual styles — and its edits land in the canvas immediately.
- Breadth of control. 40+ tools span both reading and mutating: text, fills and styling, components and instances, prototyping connections, and native annotations, so one session can audit a file and then restructure it.
- Built for scale. Bulk text replacement and instance-override propagation are chunked to survive large files, where editing nodes one at a time would stall — the gap between a demo and renaming 200 layers in a single pass.
Who It's For
Great fit if you live in Cursor or Claude Code and want an agent that manipulates design directly — auditing a design system, doing bulk copy edits, or scaffolding layouts from a prompt. Look elsewhere if you want a no-code visual editor, expect pixel-perfect generative design, or can't run a local WebSocket server alongside the Figma desktop plugin.
