[[figma]] has a new [[20250811104331-ai-mcp]] server. While it initially was mostly helpful for extract design context out of Figma, it can now also better interact with Figma. This means it helps with Prompt to Design, Design to Code, and Code to Design.
The MCP servers extracts structured design context and passes that context to your [[agent]]. The agent itself is still responsible for generating the final code. This is a lot better than using a screenshot, where the agent would have to guess what component corresponds to what, visually. This helps use better [[semantics]], Figma already knows the difference between an 'error' red and a 'brand' red. A screenshot does not.
The MCP server does not understand your design system by default though. Code connect, variables, and prompts all provide that information. https://developers.figma.com/docs/figma-mcp-server/mcp-vs-agent/
Currently the Figma MCP server has different tools to help:
Overall there's a lot of value, but it does make me worried about [[vendor lock in]]. To achieve best results, all of your information needs to be piped into Figma so it can be piped out. Things like Variables, Code Connect, etc. help with this. However it (currently) doesn't help the other parts of your design system. It doesn't pull from any standards, documentation, etc.
In an ideal world, I'd rather position Figma's MCP server as a way of gathering information from Figma, not the source of truth of generating code (which should use your DS MCP Server), similar to what's described in https://www.designsystemscollective.com/figma-dev-mode-mcp-is-nice-but-we-need-design-system-mcps-bb599582da3d. i.e. use Figma's MCP to extract design context, but not much else.
https://www.figma.com/blog/introducing-figmas-dev-mode-mcp-server/ https://www.figma.com/blog/design-systems-ai-mcp/ https://help.figma.com/hc/en-us/articles/32132100833559-Guide-to-the-Dev-Mode-MCP-Server