Difference in quality between raw Figma API and MCP output

[wrenzi]

Hi team 👋

First of all, thanks for open-sourcing this — it's been really helpful!

I’ve been testing Figma MCP with a basic login page and noticed something unexpected: the raw Figma API response produced better results than the MCP-translated version, at least in this simple case.

I understand that MCP aims to reduce payload size and improve interpretability (especially with larger files), but I'm curious — are there scenarios where the raw output might actually perform better with some models?

Also:
Aside from reducing size, what other transformations or optimizations does MCP apply to the data?
Would love to better understand what gets simplified or removed during the translation process.

I've attached the Figma file and both outputs (raw + MCP) for reference in case it's helpful.

Thanks in advance!

Figma:

Raw Figma Data:

MCP Data:

[GLips]

Weird. The translation layer isn't complete—there are certain things that aren't included in the final payload that I'd still like to add. That said, the design you shared is pretty simple. Everything should be provided.

What editor are you using, and what model produced these outputs? Are you able to share a link to the design you're working on? A minimal reproduction would be helpful—a Figma file with just this login screen.