How does Enrichr take
a site apart?
We read the rendered DOM.
Not the source you wish had shipped — the DOM after JavaScript ran and the page settled. Every node, every applied rule, every computed value.
We capture the motion, the part a screenshot always loses.
Keyframes, easings, durations, scroll triggers, hover and focus states — recorded as values you can rebuild from.
We reverse-engineer the WebGL, not just render it.
Shader source, uniforms, draw calls, and the full scene graph — geometry, materials, lights, cameras — recovered as data you can rebuild the scene from, not a video of it playing.
We extract the design system underneath the page.
Colors, the real type scale, spacing units, radii — pulled from computed styles, not guessed from a screenshot, and de-duplicated into the token set the page is actually built on.
We recover the source maps most tools skip.
When a site ships them, minified code decompresses back to original file names, real variable names, and readable function bodies — the difference between reading a rebuild and reading someone else's project.
You never start from a blank file.
We hand it straight to ChatGPT.
Design tokens, a used-CSS sheet, responsive deltas and a rebuild brief — delivered over MCP, ready to build against.
Three steps
to get started
No card required — 25 free credits every month. This is the same account and credit balance as the Claude version.
Sign in or create an accountSettings → Connectors → Advanced settings → turn on Developer mode, then Connectors → Create. Paste the URL below, and when it asks how to sign in, choose OAuth — ChatGPT will bring you back here to sign in and approve the connection.
https://chatgpt.enrichr.dev/mcp
In ChatGPT: "Reverse-engineer https://example.com"
That's it. Start analyzing.
Try NowSkip retyping it.
Copy the connector URL.
Same URL as Step 2 above — paste it straight into ChatGPT's "Create connector" dialog.
Featured Analyses
What Enrichr Detects
DOM & Structure
- Full rendered element tree
- Auto-detected tech stack
- Video, audio & media elements
- Links, images & heading counts
Animations
- GSAP & ScrollTrigger
- Framer Motion
- Lenis smooth scroll
- Exact timings & easing curves
WebGL & 3D
- Three.js & Babylon.js
- Custom GLSL shaders + complexity ratings
- 3D models (GLB, GLTF, OBJ)
- Particle systems
Deep Site Profile
- Design tokens — colors, type, spacing
- Source maps & original source code
- Fonts, frameworks & asset inventory
- Responsive ladder + interaction states
Coming next: post-processing chains, more animation libraries, performance metrics.
What a fetch cannot see
A model without a browser gets one thing: the bytes the server returns. Here is the same page measured both ways — a three.js demo whose entire content is built at runtime, in GPU memory.
And on a production marketing site — stripe.com, which is server-rendered, so the markup gap is small and the capability gap is not:
Then download the whole thing: rendered DOM, every asset pulled local, the CSS that actually applied, and the analysis — a folder that opens offline with the network disconnected.
Simple credit pricing
One credit is one ordinary extraction. A deep capture costs more because it holds a real browser for longer — a full rebuild brief is 5, a design-token read is 1, checking vision availability is free.
Pro
- 2,500 credits / month
- Vision analysis
- Priority queue · 5 concurrent
Need more for one busy month? Buy credit packs from your account — 250 for $12, 1,000 for $40, 5,000 for $175. Purchased credits never expire.
Got
questions?
-
No. You ask ChatGPT in plain English — "reverse-engineer this site" — and Enrichr does the browser automation and extraction. ChatGPT reads the results and explains them back to you.
-
A fetch gets the bytes the server returned before any JavaScript ran. Enrichr renders the page in a real browser first, so it sees what actually got applied — animations, WebGL, hover states, the CSS that survived, all of it. See the Proof section above for real numbers.
-
Anything a real browser can render — React, Vue, and Next apps, Three.js and WebGL scenes, GSAP-heavy sites, pages behind a Cloudflare challenge. If it loads in Chrome, Enrichr can capture it.
-
One credit is one ordinary extraction. Deeper captures cost more because they hold a real browser open longer — a full rebuild brief is 5 credits, a design-token read is 1, checking vision availability is free. Full breakdown in the Pricing section.
-
We store a hash of your connection's credentials, never the plaintext. Every extraction runs in an isolated browser instance, and nothing about the sites you analyze is shared with anyone beyond the ChatGPT conversation that requested it.
-
Yes — run Enrichr locally as a stdio MCP server and there's no API key or account involved at all. Accounts and credits only come into play for the hosted version.