The Agentic SEO MCP server
— a full SEO team for Claude

MCP stands for Model Context Protocol — an open standard from Anthropic that lets an AI assistant like Claude connect to outside tools and data through a server. An "SEO MCP" is that protocol applied to search work: an SEO MCP server gives the model live access to things like Google Search Console, SERP and keyword data, and a task board, so you can run analysis and fixes from a normal chat instead of stitching together APIs by hand.

Yes. You can add our SEO MCP server at https://myagenticseo.com/api/mcp for free. On first connect you generate a free professional SEO report on your real Search Console data, and as a prospect you keep read-only access to the data tools and light chat. The full research and write tools (live DataForSEO SERP, keyword, competitor and backlink research, plus the closed find-fix-verify loop) are client tier, set up after a call.

For the claude.ai custom-connector path, yes — connectors run from Anthropic’s cloud and require a paid claude.ai plan (Pro, Max, Team, or Enterprise). The Claude Code path works the same way and is aimed at developers who want the agent to edit their repo. Either way, our side is free: adding the connector and generating your report costs nothing.

An SEO API is a set of endpoints you call from your own code — you write the integration, handle auth, and build the workflow. An SEO MCP server is consumed directly by an AI assistant: the model discovers the tools, calls them in plain language, and chains them itself. Ours goes a step further than most by also handing the agent a built-in playbook, so it doesn’t just get data — it gets the workflows for how to use it: finding winnable opportunities, writing and rewriting articles that rank, and proving each fix against your Search Console.

It depends on what you need. Read-only data pipes — DataForSEO, the SEMrush and Ahrefs MCPs, GSC servers like mcp-gsc — are great when you only want to pull numbers. Ours is built for doing the work, not just reading it: it adds Google Search Console performance, a task board as the system of record, a built-in playbook that teaches the method (find winnable opportunities, write and rewrite articles that rank), repo-aware PR-only fixes, and outcomes verified against Search Console. If you want an agent that closes the loop rather than another data feed, that combination is the differentiator.

Only through pull requests you approve, and only on a client engagement. There is a human-review gate between find and fix: the agent presents ranked findings, you pick which to fix, and every change arrives as a pull request on its own branch — nothing pushes to main. Prospects connecting over the connector get read-mostly access; the only writes are task-board updates. Your site never changes without a human merging it.

Yes — the connection is deliberately narrow. At onboarding we mint a scoped token that reaches exactly one project: that project's task board, Search Console summaries, and crawled site data, nothing else. It is revocable at any time and expires on its own. Access is read-mostly — the only things the agent writes are task updates and pull requests in your repo. Your Google credentials never leave our servers, and nothing ever pushes to main.

A git-based site or repo and a normal dev workflow — someone who can review and merge a pull request. We handle everything else during onboarding. And if you don't run a coding agent in your repo yourself, that's fine: we run the same loop from our side, and the pull requests still arrive for your team to review.

Never. The agent works PR-only — every change arrives as a pull request on its own branch, and your team reviews and merges on your terms. If a PR is wrong, you close it and the task goes back on the board. Nothing reaches production without a human approving it.

SEO moves in weeks, not minutes — and the loop is built around that honestly. Fixes ship as fast as your team reviews pull requests, but Google needs time to recrawl and re-rank. That's why each fix is verified against Search Console about three weeks after it ships and marked confirmed, inconclusive, or regressed. You see real verdicts, not dashboards that assume every fix worked.

There's no public pricing because every engagement is scoped to the site. The shape is always the same — we call it the Managed Loop: a one-time setup, then a monthly retainer scoped on a call, no annual contract. Organizations that need the loop deployed inside their own infrastructure get an individually scoped Enterprise Deploy. The path starts the same for everyone: get the free SEO report on your real Search Console data, then book a call.

SEO tools stop at the list — they find issues and leave the fixing to you. AI writing tools stop at words — they produce content but never touch the technical work or check whether it ranked. This integration ships the change as a pull request in your repo and then verifies the outcome against your real Search Console data. The deliverable is a merged fix with a verdict, not a recommendation.

No — you arguably get more visibility than you have today. Every finding, every fix, and every verification verdict lives on the task board. Every change arrives as a pull request your team reviews before it ships. The connection is scoped to one project and revocable at any time. You stay the editor; the loop does the legwork.