Grow your website with real data.
The Web Anatomy MCP gives your AI agent 3,500 scored sections, 500 pages, and 4 opinionated skills, from the best websites, so it ships what works instead of guessing.
User: Improve my landing page.
- Score: 65 / 100
- 10 fixes, ranked by priority
- hero · pricing · social-proof flagged
Agent: Your page ranks 65 out of 100. Here are the 10 fixes to make, ranked by priority.
User: Show me strong signup-form sections.
- 8 signup-form sections
- Stripe · Linear · Vercel · Ramp …
- each with a scored screenshot
Agent: Pulled the top-tier signup forms from your scored library, grouped by pattern, with what makes each one convert.
User: How does my pricing page stack up?
- scored vs 24 best-in-class pages
- 3 HIGH · 5 MEDIUM gaps
- plan-clarity · qualification flagged
Agent: You trail best-in-class on plan qualification. Here is the benchmark set and what to fix first.
Your agent is guessing. It shouldn't have to.
Most AI pages are averaged from training data. Web Anatomy grounds your agent in what the best websites actually shipped.
Without Web Anatomy
- Your agent pattern-matches from training data
- Generic three-tier pricing, 'Get started' CTAs
- Galleries show pretty pages, never what works
> design our pricing page
// generic, averaged from training data
With Web Anatomy
- Ask in plain English: 'design our pricing page'
- Your agent calls a named move from a real company
- It ships what performed, with real screenshots
Design our pricing page.
Based on best practices for your industry and ICP, it pulls the pricing cards that performed:
Highlighted "Business" card anchors the recommended plan.
+ Raycast, Clay
Connect to Claude, Cursor, or any MCP-compatible client
Install MCP for freeTWO PARTS YOU INSTALL. TWO THINGS THEY UNLOCK.
Your agent audits your page and benchmarks it against the best. Here is what that looks like.
Audit straight from your code
Install the skills and your agent audits the page straight from your code. It hands back the problems to fix and their prioritization, grounded in named plays from real companies. No URL needed.
Works in Claude, Cursor, Lovable, v0, and more. Free and open, no signup.
Pull the real move
The MCP server connects your agent to the benchmark library: 3,500 scored sections from 290+ companies, with real screenshots. Your agent pulls the actual move on demand, instead of quoting a recipe.
- pricing · Wiz
- hero · Linear
- signup-form · Stripe
- social-proof · Ramp
Get a page scoring
Point it at your URL. 63 criteria scored, the gaps flagged by severity, and paste-ready rewrites grounded in the named-company sections that beat yours.
7 HIGH issues detected
Stack up against the best
Score your live URL against best-in-class pages in your industry. Get a gap report ranked by what to fix first, and get inspired by their structure and copywriting.
The move, applied to your product
It runs where your context lives, so the move comes back rewritten for you, not handed over as a screenshot.
Curated and scored, not scraped
Built by hand and tied to companies that actually performed. Curated, analyzed and scored by Gabriel Amzallag
Current, while training data is frozen
Updated weekly from real launches. Your agent's training froze months ago. This is what works today.
Free for launch. Or apply it for you.
The skills and the library are free and open forever. The MCP server is free in private beta, request a token. The audit tool exists for when you want the methodology applied to a specific page on demand.
The skills your AI agent installs to ship websites that work.
- 4 skills
- 3,500 scored sections from 290+ companies
- Skills pack for Claude, Cursor, Lovable…
- MCP server for live market data (free private beta)
- Public methodology + scoring rubric
The public reference if you'd rather read than install.
- Browsable skills library
- 3,500 scored sections, filterable by industry & section
- Real screenshots of every section
- Public methodology + scoring rubric
Real data → grounded agents → sites that grow.
Raphael Zerbib
Founder, Labonnebrique
“We help service businesses digitalize. Web Anatomy is part of our DNA now. Every project starts grounded in the library. It helps us ship faster and give clients concrete improvement ideas from day one.”
Quentin Kozyra
CMO, Mobile Club
“At Mobile Club, growth is everything. Web Anatomy helps me spot exactly where our pages are leaking conversions. The section-by-section breakdown saves me hours of guesswork and lets me focus on what actually moves the needle.”
Olivier Dumas
Founder, Stanza
“I'm a dev, not a marketer. I pull the moves from Web Anatomy and paste them straight into Claude Code. Simple fixes implemented in minutes. No design skills needed.”
Karèle Samalens
Founder, Eklo
“I run a design agency. Before every client pitch, I check their site against the Web Anatomy library. It gives me concrete quick wins to show them and proves why the redesign will actually move the needle. It's part of how we sell.”
Jean-François
Growth Manager, Pretto
“In performance marketing, you need fresh test ideas constantly. I dig through the Web Anatomy library to spot new copywriting angles, hero layouts, and section improvements I haven't tried yet.”
How to install Web Anatomy: 3 steps
Install the skills, one command, free and instant. Your agent learns the named plays.
Connect the MCP, paste one prompt or add it manually. Your agent gets live market data.
Ask, then ship, your agent calls the right skill and pulls the real move a top company performed.
Steal from the best
A scored library of real sections and industries from the top companies. Browse the patterns worth copying for what you are building.
WEB ANATOMY: FREQUENTLY ASKED QUESTIONS
What it is, how it installs, what it costs (free), and where the data comes from.
What is Web Anatomy?
[01]Web Anatomy is an open skills pack for AI agents, built from the moves real companies actually shipped. Install it into your agent, Claude, Cursor, Lovable, or anything that speaks MCP, and instead of designing your site from training-data averages, it reaches for named plays that performed for companies like Ramp, Wiz, and Figma. It comes in two parts, one install: the skills carry the judgment (what to ship, when, and why), and the MCP server connects your agent to the live library of 3,500 scored sections from 290+ companies. The skills and the public library are free forever.
Which AI agents does it work with?
[02]Any agent that supports skills, rules, or MCP. The skills pack works as Claude Skills, Cursor rules, and Lovable context, and the MCP server connects ChatGPT, Claude, Perplexity, v0, Webflow AI, Framer AI, Bolt, and Replit Agent to the live library. Per-agent install instructions live in the install guide.
Is it really free?
[03]Yes. The skills, the public reference site, the methodology pages, and the scoring rubric are free forever. No signup, no paywall. The MCP server (the live link your agent uses to query the library) is also free, but it runs in private beta: you request access, we approve it (usually within the hour), then you generate a token. That keeps the live endpoint stable while we scale it. We also offer Landing Page Analyzer, a paid auxiliary audit tool that applies the methodology to a specific URL on demand. It exists for people who don't want to install skills locally. It is not the brand; the open library is.
Where does the data come from?
[04]From 290+ companies that shipped the moves themselves: Ramp, Wiz, Figma, Notion, Loom, Webflow, Qonto, Mercury, Lovable, Clay, Lemlist, Folk, Hyperline, Speechify, ChatSEO, ElevenLabs, Navattic, Ledger, NerdWallet, Zapier, Ahrefs, Freshdesk, and many more. Their pages have been broken into 3,500 sections, each scored against a public rubric. The scoring methodology lives at /methodology, so anyone can audit how a section earned its score.
Which sections does the library cover?
[05]Every section type a real landing page ships. The 3,500 scored sections are organized by category: hero, pricing, trust signals, testimonials, CTA, FAQ, comparison, how it works, features, and more. Your AI agent picks the move that fits the section it's working on. Or browse all sections manually to study the moves before installing.
Can I contribute to the library?
[06]Yes. The skills pack at github.com/GabrielAmz/web-anatomy takes external pull requests. Spotted a play we missed? Open a PR with the strategy slug, the named company, the rule, the triage, and a screenshot. The maintainer reviews against the methodology rubric and merges if it qualifies as canonical.