Skip to content

MCP install

Web Anatomy MCP

Web Anatomy comes in two parts: free skills your agent installs, and a private-beta MCP server that gives it live market data. Add the skills in one command; connect the MCP when you want real examples and URL benchmarks.

Streamable HTTP · Bearer token · 5 workflows · webanatomy.ai/mcp

What you install

Two parts. Skills are free and instant. The MCP adds live data.

Start with the skills, no signup and no token. Add the MCP server when you want your agent pulling real screenshots, scored sections, and live URL comparisons from the library.

Free · Open · instant

1 · The skills

Named plays your agent calls. They work on their own, with no account and no token.

Run in your project

npx skills add GabrielAmz/web-anatomy

Works in Claude, Cursor, Lovable, v0, and any agent that installs skills.

Free · private beta

2 · The MCP (live market data)

A live link to the benchmark library so your agent can search 3,500 scored sections, fetch real screenshots, and benchmark a live URL on demand.

Private beta: request access in the dashboard, we approve within the hour, then you generate a token.

Install MCP for free

Request access, generate a token, and copy your guided prompt or manual config, all in the dashboard.

Add to your IDE

One config per IDE, ready to paste

Pick your IDE or agent and copy its MCP config. The snippet holds a token placeholder: generate your private beta token from the dashboard and paste it in.

  1. Open a terminal anywhere. The --scope user flag stores the server (and token) in your user config, available in all projects and never committed.
  2. Paste and run the command below.
  3. Open Claude Code and run /mcp to check that webanatomy is connected.

Where it goes: Terminal → claude mcp add (user scope)

claude mcp add --scope user --transport http webanatomy https://www.webanatomy.ai/mcp --header "Authorization: Bearer <paste-your-private-beta-token-here>"

Use the user-level (global) config so the token is never committed.

Get your token

Access flow

Four steps from request to working MCP.

  1. 01

    Ask for access

    Open the MCP dashboard, sign in, and request private beta access for your account.

  2. 02

    We review it within the hour

    A Web Anatomy admin checks the request. Once approved, token generation unlocks in your dashboard.

  3. 03

    Copy your install prompt

    Generate the token once, then copy one install prompt customized with your private token and MCP endpoint.

  4. 04

    Use sections from your agent

    Paste the prompt into Codex, Claude, Cursor, or another MCP agent to search benchmark sections and fetch examples from Web Anatomy.

Use cases

What your agent can do with Web Anatomy.

Use the MCP when your agent needs real website evidence before it recommends, edits, or writes.

01

Find proven section examples

Ask for hero, pricing, FAQ, comparison, or signup examples filtered by industry and locale.

Example prompt

Find AI hero examples with strong social proof.

02

Improve a page section

Paste your section or screenshot and ask your agent to compare it against stronger examples before editing copy or layout.

Example prompt

Improve this pricing table using Web Anatomy examples.

03

Benchmark a URL or screenshot

Give your agent a live page or screenshot and ask for gaps, stronger references, and prioritized fixes.

Example prompt

Benchmark our homepage against top AI landing pages.

04

Turn references into a report

Ask for the output in chat, a Markdown file, or an HTML report so your team can act on it.

Example prompt

Create a short .md report with examples and fixes.

FAQ

Security, access, and setup.

The dashboard handles access and token generation; your agent handles local setup from the copied prompt.

Where do I get my token?

After approval, generate it in the MCP dashboard. The install prompt is customized with that one-time token.

Where should the token be stored?

Keep it in ignored local config, such as an env var or $HOME/.webanatomy/webanatomy_mcp_token. Do not commit it.

What data can the MCP return?

It returns public-safe benchmark sections, filters, and selected section details. It does not return raw scores, marker coordinates, service-role access, or internal database fields.

Which agents does it work with?

Use the copied prompt with Codex, Claude, Cursor, or any coding agent that can install skills and configure an MCP server through mcp-remote.

What if I already use Mobbin or Lazyweb?

Keep using visual libraries for browsing. Web Anatomy is for grounding your coding agent in benchmark-backed examples it can call while working.

Ready to connect

Open the dashboard and request access.

Use the MCP dashboard to request private beta access, generate your token, and copy the install prompt for your coding agent.

Open MCP dashboard