Your AI coding agent can query 224 real products' decoded design systems over MCP: exact colors, type scales, spacing, tokens, plus 2,500+ labeled sections. It reads the real spec before it writes code. Free Chrome extension + Figma plugin included.
No reviews yetBe the first to leave a review for Mozaika
Maker
📌
Hey Product Hunt, I'm Sezai. Solo founder, building this in public.
I build everything with AI coding agents now. The code is fine. The UI always comes out the same: purple gradient, Inter, three identical cards. The agent isn't bad at design, it's blind. It has never seen the system behind the products it's imitating: the real palette, the type scale, the spacing, the button recipes.
So I spent months decoding exactly that. Mozaika is 224 real products: 295 pages, 2,500+ hand-checked sections, 242 decoded design systems, 166 of them measured from the live DOM rather than eyeballed from screenshots. You connect the MCP server once, and mid-build your agent calls get_design_system("linear") and gets structured tokens back. It reads the spec before it writes the code.
Around it I shipped a free Chrome extension (one click on any live site gives you its real design system, paste-ready for your agent), a free Figma plugin (import any product's system in one click), and free browsing of the whole library. No paywall to look.
The Founder license is $59 one-time, lifetime. First 100 seats, then $199. The counter on the site is real. I'd rather be honest than clever.
Fair disclosure: Mobbin and Refero exist, and they're great for humans to look at. Mozaika is built for the agent to consume. That's the bet, and it's unproven, which is why I'm here.
I'm at the stage where every single user matters. If you connect it to Claude Code or Cursor and run one real build, tell me what breaks. I'll be in this thread all day and I ship fixes same-day.
Question for you: which product's design system should I decode next?
Report
How fresh is the design system data when a site pushes an update, and does the MCP query return diffs or just the latest snapshot?
Report
Maker
@berralkmenz4kx great question — honest answer: latest snapshot, not diffs (yet). Every decode is dated (you'll see "Decoded Jul 4" etc. right on the page), and the values are probed from the live DOM + measured from retina captures taken that day. Flagships get re-captured regularly, and if you spot a drifted value I re-decode it same-day. Diffs are a natural next MCP tool since we already store dated versions — "what changed since my last build" is on the roadmap. Fun proof the library moves fast: the description above says 224 products / 2,500 sections — it's already 340 products / 3,600+ sections, all measured, since I scheduled this launch.
Report
Finally something that gives my AI agent actual design tokens instead of guessing from screenshots. The Chrome extension was easy to drop in and pulling a real type scale in seconds felt almost unfair.
Report
Maker
@ayselbac7079 "almost unfair" might be my favorite review ever — that's exactly the feeling I was chasing. The extension stays free forever, no account. Since you've got the type scale working: which product should I decode next for the library? Taking requests in this thread all day.
How fresh is the design system data when a site pushes an update, and does the MCP query return diffs or just the latest snapshot?
@berralkmenz4kx great question — honest answer: latest snapshot, not diffs (yet). Every decode is dated (you'll see "Decoded Jul 4" etc. right on the page), and the values are probed from the live DOM + measured from retina captures taken that day. Flagships get re-captured regularly, and if you spot a drifted value I re-decode it same-day. Diffs are a natural next MCP tool since we already store dated versions — "what changed since my last build" is on the roadmap. Fun proof the library moves fast: the description above says 224 products / 2,500 sections — it's already 340 products / 3,600+ sections, all measured, since I scheduled this launch.
Finally something that gives my AI agent actual design tokens instead of guessing from screenshots. The Chrome extension was easy to drop in and pulling a real type scale in seconds felt almost unfair.
@ayselbac7079 "almost unfair" might be my favorite review ever — that's exactly the feeling I was chasing. The extension stays free forever, no account. Since you've got the type scale working: which product should I decode next for the library? Taking requests in this thread all day.