MCP Integration — The SEO Loop in Your Tech Stack

What the MCP Integration Is

Every SEO tool — including our own free report — ends at the same place: a list of things you should fix. The MCP integration removes that last mile. It connects an AI coding agent running inside your repository to your Agentic SEO project, so findings stop being to-do items and start shipping as pull requests that get verified against your real Search Console data.

The division of labor is simple. Our platform is the brain: it finds issues, prioritizes them on your task board, and verifies outcomes after fixes ship. The connected agent in your repo is the hands: it reads the board, implements fixes in your actual codebase, and reports back. Neither half is the product. The closed loop between them is.

MCP (Model Context Protocol) is an open standard for connecting AI agents to external tools and data sources. We run a remote MCP server, and once you have your free SEO report you can connect your repository's coding agent — Claude Code, for example — to your project yourself, with one command and a sign-in. From that point on, the agent working in your codebase sees the same tasks, the same Search Console data, and the same crawl context as the agent you chat with here.

What the Connected Agent Can Do

The connection exposes 18 tools across six permission scopes, plus the orchestration context most SEO MCP servers leave out. Everything is scoped to one project, and which tools appear depends on your access level — read security and scoping for how that resolves:

Beyond tools, the server ships prompts — guided workflows the agent can invoke: find_opportunities (market-wide opportunity discovery, clustered by SERP overlap and scored for winnability), write_article, and improve_article — backed by playbook:// resources that carry the actual methodology. That orchestration layer is what separates this from a read-only data pipe.

When it runs inside your repository, the agent reads and edits the real source files and opens pull requests; new issues it discovers — a missing canonical, a broken template — go onto the same board. Note what is not on the list: the agent cannot push to your main branch, cannot touch other projects, and cannot see your Google credentials.

How the Loop Runs End to End

The loop has three phases, and then it repeats:

1. Find. The platform's weekly monitor analyzes your Search Console data and crawl results, creates new tasks, reprioritizes existing ones, and resolves what has been fixed. This runs whether or not anyone asks.
2. Fix. On a schedule, the agent in your repository picks up open tasks, implements the changes, and opens pull requests. Your team reviews and merges — a human approves every change before it reaches production.
3. Verify. Roughly three weeks after a fix ships — long enough for Google to recrawl and for Search Console data to settle — the platform compares performance before and after. Each task gets a verdict: confirmed (the metrics moved the right way), inconclusive (no clear signal yet), or regressed (the change hurt, and the task is flagged for follow-up).

The three-week window is deliberate. SEO feedback arrives in weeks, not minutes, which is exactly why scheduled agent runs are the right architecture: the loop wakes up when Google has new signal, not when a dashboard refreshes.

The Task Board as Shared State

Both halves of the loop read and write the same task board. The brain creates and verifies tasks; the hands implement them and attach evidence. On your Tasks page, completed tasks carry verification badges and links to the pull requests that closed them — so you can trace any finding from detection to merged fix to measured outcome.

Security and Scoping

The connection is designed to be narrow:

• One project per token. The connection is bound to a single client and a single project. It cannot reach anything else on the platform.
• Capability scopes, resolved live from your tier. The agent only sees the tools your access level allows, and scopes resolve from your current tier at request time. A prospect connection is read-only; when an account becomes a client, the write and market-research tools light up automatically — no re-consent needed. A prospect never widens beyond read-only.
• Expiry and revocation. Tokens expire, and we can revoke a connection at any time without touching your infrastructure.
• Hashed at rest. Tokens are stored as SHA-256 hashes on our side — the plaintext exists only in your environment.
• Google credentials never leave our servers. The agent in your repo reads Search Console data through our platform. Your OAuth tokens are never shared with your infrastructure or the connected agent.
• PR-only writes. The agent never pushes to your main branch. Every change arrives as a pull request that your team reviews.

How to Get Connected

The connection is self-serve — no tokens to copy, no configuration files to fill in. There are two paths, and the full step-by-step setup guide walks through both with every field value:

1. Add it as a claude.ai custom connector (no terminal). On claude.ai, go to Settings → Connectors → Add custom connector, name it Agentic SEO, set the URL to https://myagenticseo.com/api/mcp, skip Advanced settings (the server supports Dynamic Client Registration, so there's no client id or secret), then Add → Connect → sign in with Google → Allow. This is the path to use when you want to chat with your SEO data; it needs a paid claude.ai plan.
2. Or add the server in Claude Code if you want the agent to edit your repository. In your repo, run:

   claude mcp add --transport http agentic-seo https://myagenticseo.com/api/mcp

   Then run /mcp, choose agentic-seo, and pick Authenticate. Sign in with the same Google account, choose your project, and approve.

The first time you authenticate, if you have no free report yet, you hit the report-gate: you generate your free SEO report on your real Search Console data before the connector finishes. That report is where every engagement starts — and the same data the connected agent works from afterward.

Out of the box a prospect connection is read-only: the agent can see the task board, Search Console data, and crawl context. Write-back and the market-research tools that feed the verified find → fix → verify loop are part of the client engagement, and they switch on automatically once an account is onboarded — no reconfiguration. Book a call and the founder scopes it with you.

This page is the end-to-end deep dive on how the loop runs in your stack. For the “what is it / how do I add it” questions, defer to the canonical pages it links above: the SEO MCP server overview (what it is and how to add the connector), the explainer on why the loop matters more than the protocol, and the step-by-step setup guide. For the broader picture of how the agent reasons and chains tools, see Understanding the Agent Loop.