I kept opening Awwwards winners and seeing motion I couldn't copy. Scrapers hand you HTML and CSS. They can't see the GSAP timelines, the scroll choreography, or the WebGL shaders doing the real work. You get a corpse of the page and none of the recipe.
Designers were stuck screenshotting sites into folders and squinting at them, like archaeologists of their own industry. Meanwhile AI could write code but had no way of seeing how a page moves.
So I built an MCP server that loads a page in a real browser and reports what actually ran: the animation libraries, easings, scroll triggers, shaders, textures and geometry. The output is measurements rather than vibes, so Claude can work from real data instead of guessing what a site "probably" does.
The interface is Claude, not a Figma plugin. Design tools are turning into a layer over AI, so I built the extraction side for that world.
It's around 90% accurate on Awwwards-level sites, and it says so when it detects nothing rather than making something up. That's a higher standard than most of the industry manages.
It's pronounced "enricher". Free to try at enrichr.dev.
