Figma has shipped a Model Context Protocol server with a tool called use_figma, which lets AI agents read and write directly to the Figma canvas using your existing design system as the source of truth. This is not a prototype generator. It is a two-way bridge between a codebase and live design files.
The video follows a technical designer at a company called Earthling moving between the terminal and Figma in a single workflow: start a prompt in the terminal, iterate visually on the canvas, add comments and annotations, then push implementation back to the codebase. The chapter breakdown runs only 3.5 minutes, which means the core loop is tight and demonstrable, not theoretical.
What makes the full video worth watching is the implementation step at 2:53, where the output goes back into the codebase, not just into a handoff file. That direction of the loop is what most AI design tools skip. The MCP server documentation is linked separately for anyone who wants to wire this into their own toolchain.
[WATCH ON YOUTUBE →]