WebAnatomy

What the WebAnatomy MCP server does

The WebAnatomy MCP server gives landing-page workflows access to a benchmark of more than 3,500 scored sections across more than 500 pages and 290 companies, according to the repository. It is intended for agents working on SaaS and other product landing pages that need concrete references instead of broad conversion advice.

The underlying framework treats a page as a set of recognizable sections, including the hero, value proposition, trust elements, testimonials, pricing, FAQ, and CTA. Agents can use the benchmark to compare pages, investigate a specific section, identify gaps, and produce revised copy. The related skill pack also creates local reports and wireframes for these workflows.

How it works

The hosted MCP exposes benchmark searches and page or section lookups. The documented tools are search_pages, get_page, search_sections, and get_section. Connected skills use these results to bring current benchmark data and reference screenshots into their reports. Internal scores, thresholds, marker coordinates, and field names are translated into plain-language findings rather than exposed directly in user-facing output.

The repository describes four main working levels: finding example pages, auditing a complete page, researching one section, and writing replacement copy. A separate build step assembles structure and copy into a shareable wireframe. The moodboard workflow captures a page section by section and combines the current page, issues, benchmark screenshots, practices, and recommendations.

Setup and configuration

Install the skill pack with npx skills add GabrielAmz/web-anatomy, or install it as a Claude Code plugin, or clone the repository and copy its skills into .agents/skills/. Run webanatomy-setup after installation to record product details, target customers, industry, competitors, proof assets, and priority pages in .agents/webanatomy-context.md.

To use live benchmark data, sign in to the Web Anatomy dashboard and obtain a beta token. The repository provides a per-IDE configuration command from that dashboard rather than documenting a fixed environment-variable name. The MCP connection is optional; the skills can run without it using static and built-in reference guidance.

Tools and capabilities

The WebAnatomy MCP server supports:

• Searching benchmark pages and retrieving individual page records.
• Searching scored landing-page sections and retrieving section details.
• Supplying real reference examples and screenshots to agent reports.
• Supporting page comparison, full-page audits, section research, and grounded copywriting.

The skill pack writes artifacts beneath .webanatomy/, including Markdown and HTML reports, reference images, wireframes, and design-system notes. build-page uses a single exemplar design system when assembling a new or improved page.

Limitations and notes

Live benchmark access requires a beta token and sign-in through the Web Anatomy dashboard. Without the MCP connection, the skills remain usable but rely on static guidance and built-in patterns, so they do not provide the live benchmark results or reference retrieval described above.

The repository presents Web Anatomy primarily as an Agent Skills pack with an optional hosted MCP connection. It names Claude Code, OpenAI Codex, Cursor, and Windsurf as supported environments, along with agents that implement the Agent Skills specification. The documented material does not specify a fixed MCP transport or a named environment variable for configuration.