I've been using the Figma MCP for a while and wanted to share where it's been falling short for me, because I think it points at a positioning question.
Right now the MCP feels built almost entirely for Design→Code. That path does work — but in an inefficient way and in it's also the part Figma already handles well in manual operation (Dev Mode, Code Connect, Make), so it's not really where I was hoping an MCP would help me.
Â
The work I actually wanted to bring AI into is different:
- Reviewing UX copy
- Inventorying details that are painful to audit by hand — inconsistent states, missing variants, token drift, naming gaps
- Content management across many frames
- Most importantly, getting an AI to follow the logic of a design, not just its pixels
Â
A lot of this already sits beyond what Figma's built-in features and plugins can do, so it felt like the natural place for an AI layer to add something new. But in practice it's been rough: every call is slow and very token-heavy, I end up stacking multiple skills just to handle different scenarios, and the moment my task shifts even a little, it stops working. I don't get the sense of a tool that handles design data flexibly — it only cooperates when I hand it the exact shape it expects, which is the codegen shape.
So my honest question for the team: is this kind of work — design review, auditing, helping a team understand a file — something the MCP is meant to support, or is Design→Code the intended path? I'd find the MCP far more valuable if it leaned into the pragmatic, design-native side, and I'd love to know whether that's a direction you're considering. Either way, knowing the intended scope would save me a lot of fighting the tool for jobs it may not be built for.
Since our team works primarily in Traditional Chinese, this post was translated and polished with the help of AI — apologies if the tone comes across as a bit AI-generated.