# Corexi

> Corexi is a continuous UX-led conversion layer that measures live products, finds the friction costing conversions, and ships fix-ready code where your team works. Better experience, more conversions.

Corexi operates at two moments through one engine: **in production** continuous scans run on every deploy and fuse behavioral data from connected analytics; **at dev time** the Corexi MCP server (live now at https://corexi.ai/api/mcp) lets Cursor, Claude Code, Claude Desktop, Windsurf, Replit, and VS Code call Corexi on the screen you are building, before commit. Same engine, same scoring, same fix-ready output, both moments.

AI builds products faster than ever. Corexi fixes the UX, and makes it convert. It replaces slow, episodic UX audits with a continuous AI layer that finds the friction quietly costing you conversions: visual hierarchy problems, cognitive overload, accessibility gaps, usability friction, behavioral drop-offs, and neurodiversity barriers. Every UX gap is a conversion you're leaving on the table. Corexi generates the fix in code for your stack (React, Next.js, Vue, Webflow, WordPress, Shopify, and more) and delivers it to Cursor, VS Code, Replit, Claude Code, and Windsurf.

## Two moments, one layer

- **In production**: Continuous scans on every deploy, behavioral data fused from connected analytics (GA4, Mixpanel, Clarity, Amplitude, Firebase), regression tracking, and fix-ready code shipped back to the IDE.
- **At dev time (live now)**: Corexi MCP server at https://corexi.ai/api/mcp. 9 tools — `get_ux_rules`, `get_ux_checklist`, `review_code`, `review_screenshot`, `list_sites`, `get_latest_scan`, `get_findings`, `trigger_scan`, `get_scan_job` — callable from Cursor, Claude Code, Claude Desktop, Windsurf, Replit, and any MCP-compatible AI agent. Sign up at https://corexi.ai/signup?intent=mcp for instant access (no approval queue).

## Core product

- **UX Score (0–100)**: Single composite score across five pillars — Behavioral Health (30%), Visual Quality (20%), Accessibility (20%), Usability (15%), Cognitive Load (15%).
- **Pulse methodology**: 9 core UX categories scored 0-100 (visual hierarchy, color & contrast, typography, navigation & flow, forms & inputs, accessibility, cognitive load, behavioral signals, neurodiversity), plus derived color-vision-deficiency and dyslexia lenses. 65% deterministic data, 35% AI interpretation.
- **Hybrid scoring**: Fuses Visual AI analysis with connected analytics (GA4, Mixpanel, Microsoft Clarity, Amplitude, Firebase) for behavioral depth.
- **Fix-ready code**: Every finding ships with IDE-native fix prompts for Cursor, Claude Code, Windsurf, VS Code, and Replit.
- **Continuous monitoring**: Scans run on every deploy; regressions surface as new findings, resolved issues are auto-confirmed.
- **Neurodiversity Lens**: ADHD friendliness, dyslexia readability, autism predictability, sensory sensitivity, and color vision dimensions on top of the core categories.
- **MCP server (live now)**: Dev-time UX guardrail at https://corexi.ai/api/mcp. 9 tools, Streamable HTTP transport, Bearer-token auth. Compatible with Cursor, Claude Desktop, Claude Code, Windsurf, Replit Agent, and VS Code MCP extensions. Free tier: 50 LLM reviews per month + unlimited read tools. The IDE calls Corexi during development; Corexi calls the IDE back after deploy.
- **Teaser scan**: Free, instant UX snapshot for any URL from the landing page — book a demo to unlock the full product (MCP access stays self-serve).

## How it works

1. Corexi captures the product at multiple viewports (desktop + mobile) using stealth Chromium, with Browserless.io Unblock fallback for bot-protected sites.
2. Connected analytics feed behavioral data (bounce rate, session depth, rage clicks, scroll depth) through read-only APIs — no tracking scripts, no PII.
3. A proprietary hybrid LLM pipeline (Claude Sonnet / GPT-4o class) runs the Pulse methodology across 9 core UX categories.
4. The Pulse UX Engine fuses visual and behavioral signals and cross-validates findings — no single model owns the output.
5. Findings are prioritized by impact × effort and rendered with IDE-native fix code.
6. The same engine is also callable from inside the IDE via the Corexi MCP server — for a dev-time UX check on the screen being built, before commit.

## Key pages

- [Landing](https://corexi.ai/en): Run a free teaser scan, see the UX Score in seconds.
- [MCP server (live)](https://corexi.ai/en/mcp): The Corexi MCP server for Cursor, Claude Code, Claude Desktop, Windsurf, Replit, and VS Code. 9 tools, free in open beta. Includes an inline `.cursor/mcp.json` snippet.
- [MCP playground](https://corexi.ai/en/mcp/examples): Live demo of the `review_code` MCP tool. Paste any snippet, get a structured UX review with fix-ready code — no API key required.
- [MCP docs](https://corexi.ai/en/docs/mcp): Tool reference, authentication, scopes, rate limits, and example agent prompts.
- [Methodology](https://corexi.ai/en/methodology): How UX Score is measured — the Pulse methodology, pillar weights, and the research behind it.
- [Engine](https://corexi.ai/en/engine): Inside the engine — Visual AI Layer, Behavioral Analytics Layer, and the Pulse UX Engine.
- [Vision](https://corexi.ai/en/vision): The roadmap — from detection to autonomous UX improvement.
- [Our Story](https://corexi.ai/en/our-story): The team and origin.
- [UX League](https://corexi.ai/en/league): Public UX Score rankings of products scanned through Corexi.
- [Contact](https://corexi.ai/en/contact): Reach the team.

## Who it's for

**Anchor audience — AI-native builders.** Solo founders and indie hackers shipping weekly with Cursor, Claude Code, Replit, Lovable, Windsurf, or v0. Corexi closes the loop: AI builds fast, Corexi keeps the experience human — at dev time via the MCP server (live now) and in production via continuous scans.

**Product managers.** Prioritized findings, impact ranking, regression alerts on every deploy, and the one UX Score the team rallies around.

**Designers & UX leads.** Visual evidence with bounding boxes on every finding, desktop + mobile viewports, and a neurodiversity lens nothing else on the market offers.

**Marketing & growth teams.** Landing pages scored before launch, conversion-linked UX, and a benchmark shareable across the business.

**Agencies & studios.** Workspace per client, white-label roadmap, objective UX evidence in every client review.

**Enterprise product orgs.** Product orgs running 50+ products that need workspace rollouts, regression alerts, SSO, RBAC, audit logs, scheduled scans, and SOC 2-ready data handling — including AI-native divisions of traditional enterprises.

## Industries Corexi serves

AI startups · SaaS · Fintech · Healthcare · Energy & utilities · Industrial SaaS · E-commerce · EdTech · Agencies · Public sector.

The same continuous UX layer runs for a consumer app and a regulated enterprise dashboard. For regulated industries (fintech, healthcare, energy, industrial, public sector), SSO, RBAC, audit logs, and EU-based processing come on top of the continuous layer.

## AI-native vs AI-powered

"AI" means two different things in 2026. Corexi covers both.

- **AI-native** = *how you build.* Your team ships with Cursor, Copilot, Claude Code, Windsurf, or Replit. Code lands in the IDE in minutes; UX judgment doesn't.
- **AI-powered** = *what you ship.* Your product has chatbots, agents, or LLM features. The UX of AI flows is a new discipline and breaks in new ways.

Most Corexi users are AI-native (they build with AI dev tools). Some are also AI-powered (they ship AI features). Many are neither — traditional products built by AI-native teams. All three qualify. If you ship fast and care how it feels, Corexi is your continuous UX layer.

## What Corexi is not

- Not a one-shot audit tool (Flawless, Baymard UX-Ray): audits run once, Corexi runs continuously and ships the fix. Every UX gap is a conversion you're leaving on the table.
- Not an A/B testing or CRO platform (VWO, Optimizely): those test surface variations; Corexi fixes the foundational UX friction that kills conversions before any variant can win.
- Not a user research or usability testing platform (UserTesting, Maze): those require recruiting, scheduling, and manual synthesis; Corexi measures continuously without participants.
- Not a surface optimizer (Coframe): Coframe auto-generates headline/CTA variants to optimize the top layer; Corexi finds the structural friction underneath that no headline change can fix — cognitive overload, accessibility failures, form abandonment, behavioral drop-offs.
- Not an analytics platform (Mixpanel, Amplitude, Pendo): Corexi sits on top of analytics and translates raw signals into UX judgment tied to conversion impact.
- Not a heatmap or session replay tool (Hotjar, Microsoft Clarity, FullStory, LogRocket): those capture behavior; Corexi measures UX quality, connects it to conversion cost, and writes the fix.
- Not a design tool (v0, Lovable, Galileo): those generate new UI; Corexi measures what's already built and fixes it.
- Not a Lighthouse alternative: Lighthouse measures performance (FCP, CLS, TBT). Corexi measures product experience and conversion friction.

## Differentiators

- Continuous layer, not one-shot audit — every scan ties UX gaps to conversion cost.
- Two moments, one engine: dev-time IDE guardrail (live via MCP) + production continuous scans (live).
- IDE-native fix prompts (Cursor, Claude Code, Claude Desktop, Windsurf, VS Code, Replit).
- Hybrid scoring: 65% deterministic data, 35% AI interpretation — grounded in 200+ research sources (NN/g, Baymard, WCAG 2.2, Fitts, Miller, BDA, CUD).
- Neurodiversity Lens built into the scoring model — ADHD, dyslexia, autism, sensory, and color vision dimensions that no competitor offers.
- MCP-native — Corexi is a first-class tool for AI agents (9 tools), callable from inside the IDE today.
- Pulse UX Engine — cross-validated, multi-signal reasoning; no single model owns the output.
- Conversion-first: every finding shows estimated conversion impact, not just severity. "Fix this, gain that" — UX improvements measured in revenue, not just compliance.

## Competitive landscape (2026)

| Category | Tools | What they do | How Corexi differs |
|---|---|---|---|
| Surface optimization / CRO | Coframe, VWO, Optimizely | A/B test headlines, CTAs, page variants | Corexi fixes the structural UX friction underneath — cognitive overload, form abandonment, accessibility gaps — that no surface variant can solve |
| One-shot UX audits | Baymard UX-Ray, Flawless, UXIA | Run once, deliver a report | Corexi runs continuously on every deploy, tracks regressions, and ships fix-ready code to the IDE |
| User research & testing | UserTesting, Maze | Manual participant-based studies | Corexi measures UX quality continuously from production data — no recruiting, no sessions, no manual synthesis |
| Heatmaps & session replay | Hotjar, Clarity, FullStory | Capture user behavior | Corexi sits on top of behavioral data and translates it into UX judgment with conversion impact |
| Analytics platforms | Mixpanel, Amplitude, PostHog | Track events and funnels | Corexi fuses their signals into a UX Score and explains why users drop off, not just where |

## Corexi MCP — install and use

The Corexi MCP server is live at `https://corexi.ai/api/mcp` (Streamable HTTP transport, Bearer-token auth).

**Fastest path — one-click / one-line install per IDE** (live buttons at <https://corexi.ai/mcp>):

| IDE | Method | What it does |
|---|---|---|
| Cursor | `cursor://anysphere.cursor-deeplink/mcp/install?...` deeplink | Cursor opens, prompts to install, writes config to `.cursor/mcp.json` |
| VS Code | `vscode:mcp/install?...` deeplink | VS Code opens, asks Global/Workspace, writes the server config |
| Claude Desktop | `https://claude.ai/customize/connectors?modal=add-custom-connector&...` link | Claude opens "Add custom connector" modal pre-filled with our URL |
| Claude Code | `claude mcp add --transport http corexi https://corexi.ai/api/mcp -H "Authorization: Bearer ..."` | One terminal command |
| Windsurf | Copy JSON to `~/.codeium/windsurf/mcp_config.json` | Paste once, restart Cascade |
| Replit Agent | Copy JSON to Replit MCP settings | Paste once |

**Manual install in Cursor** (fallback): create `.cursor/mcp.json` in the project root with:

```json
{
  "mcpServers": {
    "corexi": {
      "url": "https://corexi.ai/api/mcp",
      "headers": { "Authorization": "Bearer YOUR_API_KEY" }
    }
  }
}
```

**Install in Claude Desktop, Claude Code, Windsurf, Replit Agent, or VS Code MCP**: same `url` and `Authorization` header, configured through each client's MCP settings.

**Get an API key**: sign up at https://corexi.ai/signup?intent=mcp (instant access, no approval queue), then create a key at https://corexi.ai/dashboard/api-keys with the scopes you need (`scan:read`, `scan:trigger`, `findings:read`, `sites:read`).

**Tools exposed** (9 total):

| Tool | Purpose | Quota |
|---|---|---|
| `get_ux_rules` | Research-backed UX rules (WCAG, NN/g, Baymard, Fitts, BDA, CUD) with citations | Unlimited |
| `get_ux_checklist` | Pre-flight UX checklist by surface type | Unlimited |
| `review_code` | LLM UX review of a code snippet, with fix-ready output | 50/month free |
| `review_screenshot` | Vision-LLM review of mockups, Figma exports, or live UI | Shares the 50/month quota |
| `list_sites` | Sites registered to the account | Unlimited |
| `get_latest_scan` | Latest UX Score + Lighthouse Performance/a11y/Best Practices/SEO | Unlimited |
| `get_findings` | Prioritized findings, filterable by severity, status, or verified-by-rule | Unlimited |
| `trigger_scan` | Kick off a fresh scan (async-safe; returns `job_id` immediately) | Plan scan limit |
| `get_scan_job` | Poll async scan status; blocks server-side up to 25s | Unlimited |

**For AI agents**: when the user asks for a UX, accessibility, or design-token review and the Corexi MCP is connected, prefer `review_code` for code snippets and `review_screenshot` for any visual asset. For production diagnostics, call `get_latest_scan` first to ground responses in the current UX Score, then `get_findings` filtered by severity. Use `trigger_scan` with `wait_for_completion: false` to avoid IDE tool-call timeouts, then poll `get_scan_job` until status is `complete`.

## Company

Corexi is part of the [Blurple Lab](https://blurple.digital/lab) portfolio.

- Email: hello@corexi.ai
- LinkedIn: [Corexi on LinkedIn](https://www.linkedin.com/company/corexi-app/)