Figma's MCP is incredibly difficult to use. Calling Figma MCP with ClaudeCode takes an extremely long time and consumes a lot of tokens when generating design files, and the generated quality is also low. All my design systems are actually based on Figma, and now I have to explore new design software such as Apple Pencil, Claude Design, and Open Design, migrating component assets elsewhere. It's heartbreaking. Figma could definitely be better than any other design tool; it's just really hard to use right now, and in the team's exploration of AI workflows, designers are forced to migrate assets.
Question
figma mcp
Hi,
Thanks for your post. I can see how the combination of MCP slowness and credit consumption makes this difficult for you, especially if you aren’t getting outputs that meet your expectations. To help troubleshoot the MCP slowness, I’ve created a ticket for you and escalated that to our Technical Quality team. Can you watch for that email and reply with more information? Information that would help includes if you use a VPN, errors messages you might get back from the MCP, or if it is consistently slow or only sometimes, etc.
In regards to credit consumption and output quality specifically with the MCP server, are you using any of our MCP skills? You can find a full list of the Figma MCP skills in the help center. Skills give Claude Code extra guidance for tasks in Figma, and reduce the amount of prompting you need give so your results are more reliable.
I appreciate your kind words that Figma is better than other design tools, and we as a team want to make sure that it remains something you can use reliably and consistently.
Tokens getting used up are a real pain. There are a few things I have seen work across the board for me. Sizing of the prompt matters. If I try getting too much done - whole UI instead of say a component at once or a smaller chunk - the quality suffers, and usually the outcome suffers. For quality, Figma MCP gives raw node data with a full JSON tree of everything selected. That means without proper scoping, requests start to get increasingly slower and token-consuming. Plus, if the file itself is big, that makes it worse. Try setting constraints for MCP by moving everything relevant into a defined frame, and experiment for yourself to see what "right scope size" would mean for your UI elements. Getting AI to define for AI helps too, quite a lot. It's like an agent knows another agent better. Would love to know if you have already tried these or what changes they bring once you try them.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.