Paste a snippet.
See Corexi's real review.
No login. No API key. The same structured findings Cursor and Claude Code agents receive through the MCP review_code tool — rendered live in your browser.
Pick an example
Primary button uses a pale gray that fails WCAG AA on white.
5 reviews/hour per IP on this demo. The authenticated MCP server has a 50/month free tier with no per-IP cap — grab a key.
Pick an example, hit Review, and Corexi will produce structured findings.
The same structured output ships through your IDE via the MCP review_code tool — with your design tokens, monthly quota, and audit log applied.
What you just saw, in plain words
The output isn't a screenshot or a PDF. It's the same structured JSON your IDE's agent reads — so the agent can actually act on it.
11 research-backed categories
Visual hierarchy, CTA visibility, form design, spacing, typography, color & contrast, mobile, cognitive load, accessibility (WCAG 2.2), color blindness, dyslexia.
Citations, not vibes
Every finding cites a rule id from WCAG, NN/g heuristics, Baymard, Fitts, Miller, BDA, or CUD — same registry your IDE agent reads.
Fix-ready code
Each finding ships a fix_code snippet. With design tokens registered, hardcoded hex/px values are auto-replaced with your var(--token) names.
Get this in your IDE in 60 seconds.
The public demo runs at 5/hour per IP. With an API key you get the full 50 review_code calls/month, plus all the read tools and trigger_scan — no per-IP cap.