Design System Agent Readiness Checker
brought to you by zeroheight

Your documentation was written for humans.

Here’s what happens when an AI reads it.

DS ARC is a tool that helps you test the agent-readability of your design system documentation. We run a sample of your docs against a rubric based on the Design System Documentation Spec, and then point an agent at the docs with eight real jobs. Then we show you the transcripts, and give you real-world advice on how to make your docs more agent-readable.

Use thetool →

Provide your public documentation

For documentation that’s already published on the open web. We crawl the URL you give us the same way an agent would.

Documentation URLThe top-level documentation URL — the index that links to your components, tokens and foundations. Which URL you give changes what we can read, so check it below before running.Another way to reach your docs — optionalWe ask rather, than infer. If your docs are served through a channel we don’t test, what a crawler couldn’t fetch is evidence about what was exposed, not about your documentation.Include this run, anonymised, in zeroheight’s annual Design System Report and State of AI and Design Systems report. Anonymised means we keep the scores and the counts and drop everything that could identify the system — its name, URL, pages, transcripts — and only ever publish totals across many systems. Optional, and it does not change the report. How we handle your data

One run reads up to 40 pages, sampled across your sections rather than crawled in order, and takes about five minutes. Your report link is unguessable, not indexed, and expires in 14 days. There is a limit on how often one site can be checked.

Understand thetool →

Testing against a solid rubric

One of the questions we get asked a lot is ‘how do we make sure the documentation we’re writing is good for AI agents?’ We strongly believe that good documentation should stand up well for both humans and machines, but there’s possibly still some work to be done for most teams. To start with, we run a sample of your documentation against the conventions outlined in the Design System Documentation Spec, a community-run spec for machine-readability of design system documentation. We mark your docs against 12 guidelines drawn from the spec, to give you an overall ‘machine-readability’ score, as well as improvements you can make tomorrow.

Testing against a real-world agent

Running a rubric against your documentation is only theory, though. We want to show you what happens when a real AI-agent reads your documentation with queries that mimic real-world application. That’s why we formulate a series of tests based on the documentation we crawl, and run them multiple times against your documentation using Anthropic’s sonnet-4.5 model. Each test is designed to answer a specific area, from token lookups, to ambiguous components, to components that don’t exist. We then show you exactly what worked and what didn’t work, with suggestions on how to fix your documentation to ensure you’re providing the best context possible.

Hand AI agents one tool to give them all the design system context they need.

zeroheight is the documentation platform that brings the whole design system together: Figma, Storybook, tokens and guidelines on one published page, with governance built in. One MCP connection then serves that context to every AI tool you use.

See zeroheight MCP →Book a demo

Already on zeroheight? Run DS ARC on your styleguide →

What the agent asked forWhat zeroheight hands it

Which component, and when
Choosing a component · Choosing between two · Look-alike names
Usage guidance on the component page: when to use it, when not to, and the do-and-don't pairs. Neither Figma's MCP nor Storybook's carries this.
What it takes
Component props · Allowed values
Stories and code on the same page as the rules, so props and variants arrive with their constraints.
Which token and when to use it
Token lookup
Token sets that say what each token is for, not just what it is called.
How it behaves
Keyboard interactions
Accessibility guidance where the component lives.
Whether it exists at all
Invented component
Published pages only. Drafts and hidden content never reach an agent, so it cannot build from a component that isn't there.

1published sourceThe whole system on one page: Figma and Storybook plug straight in. Designs, code, tokens and guidelines sit together, with inbuilt governance. No more spreading your docs against twenty different tools.
Hoursnot monthsDocumentation to a good standard: Agentic flows draft, audit and fill the gaps in your pages against the standard this report scores, taking the work from months to hours.
10×fewer tokensOne MCP for every agent: Cursor, Claude Code, Copilot, Figma's agent, v0, Lovable and any LLM read the same published context. Enterprise teams see a 10× cut in AI token use against markdown or raw files; Free and Starter see 3×.