In June we scored every design system in our gallery with a full editorial breakdown — 37 of them — against five concrete signals of AI-readiness: a first-party MCP server, an llms.txt on the canonical docs domain, tokens in the W3C/DTCG format, a CLI-installable component registry, and published Figma Code Connect mappings. First-party evidence only: community wrappers and "works with Figma" claims didn't count. The full table lives on the Agent-Ready Index; this is what the data actually says.
10/37
ship llms.txt
11/37
ship first-party MCP
3/37
confirmed DTCG tokens
2/37
publish Code Connect
1/37
ship a CLI registry
Finding 01
llms.txt is the race everyone enters because it's free.
10 of 37 systems publish one — by far the most-adopted signal. It's also the shallowest: most llms.txt files we found are sitemap dumps, not curated entry maps. Shipping one takes an afternoon on any docs platform, which is exactly why it can't differentiate anyone anymore. If your llms.txt just mirrors your nav, an agent learns nothing it couldn't crawl.
Finding 02
First-party MCP tracks commercial pressure, not engineering maturity.
11 of 37 ship a real MCP server, and the pattern is unmistakable: they're almost all commercial or open-source component libraries — MUI, Chakra, Mantine, Ant, the shadcn ecosystem, Atlassian. Teams whose users are developers building with agents felt the pull first. The famous brand systems — Apple HIG, Google Material, Airbnb DLS, Stripe's design surface — ship none. Not because their engineers can't; because no external customer is asking. Internal pressure arrives later, when their own product teams start shipping through agents.
Finding 03
The CLI-installable registry is still a category of one.
Exactly 1 system in the audit distributes components the way agents want to consume them — npx shadcn add <url> straight into the repo, where the type-checker becomes the guardrail. Tailwind, Radix, and Geist are obvious candidates and none have shipped it. This is the widest-open opportunity on the board: the pattern is proven, the spec is public, and almost nobody has moved.
Finding 04
DTCG conformance is nearly invisible — and probably undercounted.
Only 3 of 37 systems publish tokens we could confirm in W3C format ($value / $type / $description). Most ship tokens as bespoke JSON or Style Dictionary output with no conformance claim either way — the spec only stabilized in late 2025 and many pipelines are mid-migration. We score "unknown" rather than "no" when evidence is absent, and this is the signal where that honesty matters most. Expect this column to move fastest between audits.
Finding 05
The net: docs-first React libraries lead; brand systems trail by a year or more.
The top score in the audit is 3/5 — nobody is fully agent-ready. Meanwhile 20 of 37 systems score 0/5: no signal at all, on any axis. The systems closest to ready share a shape — open-source, React-centric, docs site as the product. The systems furthest away are the ones designers admire most. If the last decade's status hierarchy ran on visual polish, the next one runs on machine-readability, and the leaderboard reshuffles accordingly.
If you own a system
Three moves, in order.
Ship a curated llms.txt this week — it's free and stops the crawl-guessing. Get your tokens into W3C format this month, with a real $description on every semantic token — that's the field agents use to choose between look-alikes. Then expose the catalog over MCP once the tokens are clean. The full sequence, with a diagram per step, is the structure guide.
The data
Every score and evidence URL is open — JSON and CSV, CC BY 4.0. Audited 2026-06-10; re-audited quarterly. If we got your system wrong, the evidence link is right there in the table — tell us and we'll re-check within the cycle.