GEO MCP Explained: What It Is, The Real Repos, How To Install One, And The Models That Drive It

A GEO MCP is a Model Context Protocol server that gives an AI agent generative engine optimisation tools, so the agent can audit pages for AI search visibility on its own.

MCP is the open standard that lets agents like Claude Code and Hermes Agent use outside tools, and GEO is the work of getting your business named in answers from Google AI, Gemini or ChatGPT.

The problem a GEO MCP solves is simple, because checking crawler access, schema and quotable passages by hand on every page takes ages.

I read the README of every real GEO MCP I could find on GitHub today, and they're all small community projects with between zero and three stars.

My number one pick is my own Agent OS, so yes, this is my own list.

If you'd like my team to look at your AI visibility with you, book a free one-to-one GEO, AEO AI audit and get a custom game plan.

Near the end, I'll show which GoldieBench-scored models I'd trust to drive a GEO MCP, because that's a separate question from the tools themselves.

The best GEO MCP options ranked

This table gives you the answer first, with my own Agent OS at the top.

I rank Agent OS first because a GEO MCP only hands your agent some checks, while the hard part of GEO is choosing what to publish and publishing it.

I haven't run the four repos myself, so everything below about them comes from their own READMEs and GitHub data I pulled on 19 October 2026.

How does a GEO MCP work?

Your agent starts with no tools in its hands, and an MCP server gives it a tool belt.

When the agent decides it needs a check, it calls a tool on the server by name, the server runs the check and the result comes back into the chat.

A GEO MCP's tools usually check robots.txt for AI crawlers like GPTBot, ClaudeBot and PerplexityBot, validate JSON-LD schema, score passages for how quotable they are and check llms.txt files.

That covers the audit half of GEO.

None of these tools can choose your next topic or give you a real case study to write about.

The four GEO MCP repos, from their own READMEs

Here are the GitHub numbers I pulled on 19 October 2026.

tmolavi/mcp-geo-server

The README describes a Python adapter by Taqi Molavi that exposes the SAGE audit engine to agents like Claude Desktop, Cursor and Antigravity.

It audits a live URL or raw HTML across technical SEO, entity AEO and GEO readiness, with extra tools for entity analysis, citation readiness and llms.txt generation.

It labels each finding from E0 for facts to E5 for hypotheses, and it says its Citation Survival Proxy is a diagnostic rather than a citation probability.

The README installs it from PyPI, but I couldn't find the package on the public PyPI index on 19 October 2026, so you may need the repo source.

jpurnell/GeoSEOMCP

The README lists 29 tools across citability, content analysis, crawler access, schema, technical SEO, brand and platform presence, llms.txt and composite scores.

It needs Swift 6.2 or newer and macOS 14 or newer, and it installs by cloning the repo and running swift build.

The README says no tool fetches a URL, so you supply the page text and the scores stay offline and repeatable.

SEOSiri-Official/aeo-geo-mcp

The README describes a local-first server by Momenul Ahmad of SEOSiri for AEO, GEO, llm.txt auditing and content stickiness analytics.

Its 10 tools include an llm.txt audit, a 0 to 100 GEO readiness score, answer-card extraction and a schema density check.

henu-wang/geoscore-mcp

The README describes a server for the hosted GEOScore platform, with eight tools, where the deep scan, the fix generator and the citation check need an API key.

The code hasn't been pushed since 5 March 2026, geoscoreai.com returned an error when I tried it on 19 October 2026, and I couldn't find the package on npm that day.

📞 Want to know which checks actually matter for your site?

Book a free AEO SEO strategy session and my team will look at your site, your competitors and the keywords you're closest to winning.

→ Get your free one-to-one GEO, AEO AI audit

How to install a GEO MCP in Claude Code or Hermes

In Claude Code, Anthropic's MCP docs use claude mcp add --transport http <name> <url> for a remote server.

For a local server, the docs use claude mcp add <name> -- <command> [args...], with the launch command from the server's own README after the double dash.

You then check it with claude mcp list, claude mcp get <name> or the /mcp command inside Claude Code.

In Hermes Agent, the Hermes MCP docs say you add the server under mcp_servers in ./.hermes/config.yaml with a command and args, or a url for a remote server.

Hermes discovers the tools at startup, and a tools: include: list under the server means only the tools you reviewed get registered.

Review any GEO MCP before you install it

An MCP server is code that runs and acts for your agent, so the review comes before the install.

Anthropic's docs say to verify you trust each server before connecting it, and they warn that servers that fetch outside content can expose you to prompt injection.

The Hermes docs say you should read what a catalogue entry installs and runs before you install it.

I'd read the source, list which tools read and which write, whitelist only the read-only tools and test on a throwaway project first.

My workflow doesn't need a GEO MCP at all

In my video above, the workflow I use to get named in AI answers runs on an agent, my Google Search Console data and an SEO skill.

I can run it from my Agent OS or from Hermes Agent directly, because it's the same process either way.

Hermes Agent is free and open source, and I use OpenCode as the free brain.

I'm honest in the video that this is Google organic data, because Google and the AI engines don't report AI traffic well yet.

My SEO skill is a checklist of roughly 100 points, and it writes everything from a case-study angle.

These are my own results from my own screens, not a promise.

For "best geo expert", my content shows in the Google AI answer at the top and ranks first organically.

In AI Mode, asking for the best GEO experts to follow in 2026 lists me first and links to my sites.

One of my sites went from zero to more than 1,300 clicks a day, and two other sites are on the same path.

Google's guidance and the llms.txt question

Google updated its page on using generative AI content on 1 October 2026.

It says it's "critical to manually factcheck and review all AI-generated content" before publishing, including titles, descriptions, structured data and alt text.

So any schema or meta tags a GEO MCP generates still need a human check.

Google's AI features page also says you don't need AI text files or special schema to appear in AI Overviews or AI Mode.

All four repos check or generate an llms.txt-style file, so I treat that score as optional for Google.

Limits of any GEO MCP

Every GEO MCP I found is a small community project, so updates and support aren't guaranteed.

A GEO score is a proxy, and no tool can tell you what an AI answer will say tomorrow.

Nobody can guarantee you'll be named in an AI answer, and that includes me.

📞 Want a plan instead of another score?

Claim a free strategy session and leave with a custom game plan to rank in Google and AI search.

→ Book the free call

Which AI models should drive a GEO MCP?

A GEO MCP doesn't ship its own model, and neither does Agent OS or Hermes Agent.

The model is the brain that decides which tool to call, reads the result and writes the fix list, so it matters more than people think.

Here's how some models you can run Hermes or Claude Code on score on their live GoldieBench pages today.

Claude Opus 5 is the highest of these at 8.27, with Hermes MoA at 8.17 and GPT-5.6 Sol at 8.16 close behind.

In my video, the free brain was DeepSeek V4.1 Flash inside OpenCode, and that exact model isn't on the board.

The closest model on the board is DeepSeek V4 Flash, and its live page lists it as currently unranked, so I can't give you a score for it.

Why build scores and GEO audits are different questions

GoldieBench scores one-shot builds, which is a different job from calling a GEO audit tool and reading the result.

A strong builder is usually good at following long instructions, which helps when the agent has to work through a skill file step by step.

A free model can still run the loop well, which is why I use a free brain in the video.

Whatever model you pick, the fact-check stays a human job.

Where the AI Profit Boardroom fits

Inside the AI Profit Boardroom you get my Agent OS, my SEO skill checklist and the AI SEO automation section.

You also get Hermes trainings that include Hermes with Claude MCP, four coaching calls a week and a member map with 3,400+ members.

Related reading on GoldieBench

My GEO examples post shows four real AI answer wins from my own screens.

My free GEO audit post shows how to check your site before you build new pages.

In the same batch, I wrote a step-by-step guide on how to do GEO for a website.

If you want expert eyes on your plan, book a free strategy session and see the keywords you're closest to winning.

Also On Our Network

• Agent Operatorsa step-by-step GEO MCP setup for Claude Code and Hermes
• AI Income Deskfive practical ways agencies use a GEO MCP
• AI Tool Verdictevery GEO MCP server rated feature by feature
• agentos.guidea GEO MCP quick start for your Agent OS
• agentos.guidemy Claude AI SEO rank #1 guide

Start with your own Search Console data, book a free strategy session if you want a second opinion, and add a reviewed GEO MCP only where it saves your agent real work.