Skip to main content

See Corexi run on your own product, live, in 15 minutes.

Book a demo
corexi
Integration · Model Context Protocol

Your IDE meets a UX consultant that ships fix-ready code.

The Corexi MCP server lets Cursor, Claude Code, Windsurf, Replit and VS Code call Corexi directly: research-backed rules while you design, a review before you commit, production findings after you ship. No browser tab, no copy and paste.

Get your API key →Try a live reviewSetup guide
  • Cursor
  • Claude Code
  • Claude Desktop
  • Windsurf
  • Replit Agent
  • VS Code
MCP tools
10
IDEs supported
6
LLM reviews free
open beta
Credit card
$0
Cursor · Composer● corexi MCP
Review this CTA before I commit. Use our design system.
<button className="bg-[#94a3b8]
  px-3 py-1.5 text-white">Subscribe</button>
✓Called review_code · 1.2s
3 FINDINGS · 1 CRITICAL
AAContrast 2.1:1 fails WCAG AA. Use var(--color-primary).
44Tap target 36px, needs 44px or more (Apple HIG, WCAG 2.5.5).
⌨No focus-visible ring. WCAG 2.4.7.
<button className="bg-[var(--color-primary)]
  min-h-11 px-4 text-white
  focus-visible:ring-2">Subscribe</button>
Two moments

Build it right, then keep it right.

Ten tools: four for the moment before the bug ships, six for after it is live.

Build time

Catch issues before they exist.

Your IDE asks Corexi for the right pattern, copy and structure before the first line of UI is written.

  • get_ux_rules

    60 research-backed rules. WCAG, NN/g, Baymard, Fitts — with citations.

  • get_ux_checklist

    Pre-flight checklist for the surface you're about to build.

  • review_codeopen beta

    Send a snippet, get findings + fix-ready code in your design tokens.

  • review_screenshotshared quota

    Vision-LLM review of a mockup or Figma export at design stage.

After deploy

Fix what production tells you.

Your IDE pulls the latest run, narrows to the issue that costs most, and lands fix-ready code in your editor.

  • list_sites

    Sites registered to your account.

  • get_latest_scan

    Latest UX Score + Lighthouse perf / a11y / SEO.

  • get_findings

    Prioritized findings, filterable by severity, status or verified-by-rule.

  • trigger_scanscan scope

    Start an Autopilot run on the live product, from the editor. It signs in and walks the product; you keep typing.

  • get_scan_job

    Ask how the run is going, or wait for it, without blocking the IDE.

  • get_run_diff

    After a deploy: what the run fixed, what is new, what came back, what was not seen — against the run before it.

What it looks like

Four prompts, four answers.

Each is a tool call or two, from design to the run that confirms the fix. Your agent picks the tool; you stay in flow.

  • Before you build

    I'm about to build a checkout form. What should I get right the first time?

    get_ux_checklist()
    • 12 surface checks
      Single-column layout, autofill attributes, error placement…
    • 8 applicable rules
      Severity-tagged: REQUIRED / RECOMMENDED / OPTIONAL.
    • Citations attached
      Each rule cites WCAG · Baymard · NN/g.
  • Before you commit

    Review this CTA button JSX before I push. Use our design system.

    review_code()
    • 3 findings
      Contrast fails AA, tap target 36px, no focus-visible ring.
    • Token-aware fix
      Fix uses var(--color-primary), var(--space-md). Your tokens.
    • Rule cited
      WCAG 1.4.3, Apple HIG tap targets, WCAG 2.4.7.
  • After you ship

    What's the highest-impact UX issue on our app right now?

    get_findings()
    • Critical · /pricing
      Monthly toggle has no aria-label — announced as 'button'.
    • Verified
      axe-core confirmed the violation. Lighthouse a11y 87/100.
    • Drop-in fix
      Ready Toggle with aria-checked + aria-label.
  • And then confirm it

    I shipped the fix. Run Autopilot again and tell me when it's done.

    trigger_scan() → get_scan_job()
    • Run started
      Autopilot signs in and walks the product; the call returns a job id straight away.
    • You keep typing
      get_scan_job() reports the steps as they happen, or waits for the end.
    • Closed the loop
      The finding resolves on the next run, and the run says so.
Setup

Connect in 60 seconds.

Cursor, VS Code and Claude Desktop install in one click. Claude Code, Windsurf and Replit get a single command or a ready config.

  1. 1
    Generate an API key

    Sign up free, then create your first key in Settings → API keys.

  2. 2
    Drop in the config

    Paste the JSON into .cursor/mcp.json or your IDE's equivalent, or use a one-click install below.

  3. 3
    Ask anything

    Try "Review this button" or "What's the latest UX score on my app?"

Buttons install Corexi with a YOUR_API_KEY placeholder. Generate a real key at Settings → API Keys (free, instant for MCP signups) and swap it in once.

  • Cursor

    One-click install

    Add to Cursor

  • VS Code

    One-click install

    Add to VS Code

  • Claude Desktop

    Opens Claude with our connector pre-filled

    Add to Claude

  • Claude Code

    One terminal command

  • Windsurf

    Paste into mcp_config.json

    ~/.codeium/windsurf/mcp_config.json

  • Replit Agent

    Paste into your Replit MCP config

Why it is safe to give to your IDE

Built for production teams.

Owner-only auth

Every call carries a Bearer key, verified server-side (SHA-256 with a per-key salt, constant-time compare). No anonymous MCP traffic.

Telemetry that respects you

We log call duration, status and structured arguments. We never store raw code snippets, images, or anything you did not ask us to keep.

Design-system aware

Pass your site_id and reviews use your registered design tokens. Fixes return var(--your-token), not hard-coded hex or px.

A free tier that stays free

Reviews are open during the beta. Read tools unlimited. trigger_scan follows your plan's monthly cap.

Stop alt-tabbing. Stay in your editor.

Generate a key in under a minute, paste the config, and your agent gets a UX consultant and the production runs, for free.

Get your API key →See the platform