We audited 193 pages with our open-source GEO MCP. Here is what it found.

We built GEOMCP, a free MIT-licensed MCP server for technical GEO and SEO checks. I’m on the WebDecoy team.

Then we ran it against all 193 URLs in our production sitemap through a real SDK client and stdio MCP server.

What needed attention

• Five articles had unnamed checkboxes generated from Markdown task lists.
• The glossary and bot report skipped a heading level.
• Our Features page linked to AI Analytics with the label “Learn More.”
• Three standalone pages had no contextual same-origin links in their main content.

We fixed the issues, deployed the changes, and ran the same MCP on those pages. All 11 original warnings cleared.

The run also found a bug in our comparison tool: it classified unchanged unknown results as coverage changes. We fixed that and added a regression test.

What the score does not tell you

182 pages initially scored 100 on the basic checklist. That does not establish AI citations, perfect accessibility, good Core Web Vitals, or higher conversion. The evidence behind each check is more useful than treating the score as a forecast.

GEOMCP has five read-only tools: audit a URL, audit supplied HTML offline, inspect robots policies, inspect a sitemap, and compare reports. No API key is required. It inspects server-delivered HTML and does not run page scripts.

See the real before/after results, the demo, and installation instructions.

Source and MIT license. Feedback and reproducible issues are welcome.

This write-up was prepared with AI assistance from recorded test output and reviewed against the implementation.