Connect the Agentic SEO MCP Server

The Agentic SEO MCP server is a remote MCP server you connect to Claude so the agent can read your Google Search Console data, run SERP and keyword research, and work a shared task board on your behalf. You can add it in two ways: as a claude.ai custom connector with no terminal, or in Claude Code if you want the agent to edit your site repository. Both paths sign you in once and stay connected.

What an SEO MCP Server Is

MCP — the Model Context Protocol — is an open standard from Anthropic for connecting AI agents to external tools and data. An MCP setup has three parts: a host (the app you talk to, like claude.ai or Claude Code), an MCP client inside it, and an MCP server that exposes tools the agent can call. An SEO MCP server is simply an MCP server whose tools are SEO data and actions — keyword research, SERP analysis, Search Console queries, and so on — so an AI agent can do SEO work in natural language instead of through a dashboard. For the SEO MCP server overview and the commercial picture, start there; for the concepts behind it — why the loop matters more than the protocol — read the explainer. This page is the how-to.

Our server lives at https://myagenticseo.com/api/mcp. It exposes 18 tools across six permission scopes, plus the orchestration context most SEO MCP servers leave out: MCP prompts (find_opportunities, write_article, improve_article) and playbook:// resources that tell the agent how to run the work, not just which APIs exist. It uses OAuth 2.1 with Dynamic Client Registration, so there is no client id or secret to paste anywhere.

Before You Start

You need three things in place:

• An Agentic SEO account. Sign in with Google — this is the same account the connector signs into.
• Google Search Console connected. The server reads your real GSC data, so connect a GSC property first.
• A way to run Claude. For Path A, a paid claude.ai plan (Pro, Max, Team, or Enterprise) — custom connectors run from Anthropic's cloud and are a paid-plan feature. For Path B, Claude Code installed in your repository.

Path A — Add as a claude.ai Custom Connector

This is the no-terminal path and the one most people should use. The connector runs from Anthropic's cloud, so the agent on claude.ai can read your SEO data directly in a chat.

1. On claude.ai, open Settings → Connectors (under Customize → Connectors), then click + / Add custom connector.
2. Fill in two fields:

   Field	Value
   Name	Agentic SEO
   URL	https://myagenticseo.com/api/mcp

   Skip Advanced settings — our server supports Dynamic Client Registration (RFC 7591), so there is no OAuth client id or secret to enter.
3. Click Add, then Connect. A window opens — sign in with Google, review what the connection can see, and Allow.

That's the whole setup. The connector now appears in your chat's tool list and stays signed in.

Path B — Add in Claude Code

Choose this path if you want the agent to edit your site repository — turn findings into pull requests in the codebase it's running in, not just read data. In your repo, run one command:

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

Then run /mcp, choose agentic-seo, and pick Authenticate. Your browser opens — sign in with the same Google account, choose your project, review the scopes, and approve. The connection stays signed in from there.

The Free Report Step

The first time you authenticate, if your account has no free report yet, you hit a report-gate instead of the consent screen. This is by design, not a blocker: you generate your free professional SEO report on your real Search Console data before the connector finishes. The report is where every engagement starts — and it's the same data the connected agent works from afterward.

So “add the MCP server” naturally produces a free report. Once the report exists, the consent screen appears, you approve, and the connection completes. You can also generate the report first and then add the connector — either order works.

What the Connection Can Do

Once connected, you ask in plain language and the agent calls the right tool. The 18 tools group by scope:

Beyond tools, the server ships prompts — guided workflows you can invoke directly, each carrying the methodology, not just the API. They map to three outcomes: find winnable opportunities (find_opportunities — market-wide discovery, clustered by SERP overlap and scored for winnability), write and rewrite articles that rank (write_article and improve_article, both run against a SERP teardown and the article evals before a draft lands), and — once a fix ships — prove the result in your Search Console through the verified loop. The playbook:// resources behind these prompts are what separate this from a read-only data pipe: the task board is the system of record, and the agent populates it as it works.

Scopes and Access Levels

The tools you see depend on your access level, and scopes resolve live from your current tier at request time — no re-consent needed when your access changes:

The serp:read tools are client-only because they trigger paid external data calls. Becoming a client is a human-closed step — book a call and the founder scopes it with you. The security model is narrow by design: one project per connection, OAuth 2.1 with PKCE, tokens stored only as SHA-256 hashes, and your Google credentials never leave our servers — the agent reads GSC through the platform.

Troubleshooting