Check website

Agentic SEO with an SEO MCP server

·4 min read

An SEO MCP server gives an AI agent such as Claude Code, Cursor or Codex a set of SEO checks it can call on its own. The agent runs a check on a live page, edits the code that caused each problem, deploys and runs the check again to confirm the fix. That loop is what people mean by agentic SEO. Before you connect one, look at how it authenticates, how it caps spending, whether retries can charge twice, and what it is allowed to change.

What is an MCP server?

An MCP server is a program that exposes tools to AI applications over the Model Context Protocol, an open standard for connecting AI apps to outside systems. The client, your coding agent, asks the server which tools it has, reads each tool's description and input schema, and calls the ones that fit the task.

The transport spec defines two standard transports. stdio runs the server as a local process. Streamable HTTP talks to a remote server over HTTPS, which is what a hosted SEO service uses. A remote server needs only a URL and a key, with nothing to install.

What agentic SEO looks like in practice

Agentic SEO means the agent does the whole loop, from finding a problem to proving the fix worked. Here is how one round goes on a pricing page, as an illustrative example.

  1. You ask: "check example.com/pricing and fix what you find."
  2. The agent calls an on-page check. It comes back with two warnings. The title is 74 characters and will truncate, and the page has no meta description.
  3. The agent opens the template that renders /pricing, shortens the title and writes a description from the page's own copy.
  4. You deploy, or the agent pushes to a preview URL.
  5. The agent runs the same check on the live URL. Both warnings are gone, so it reports the page as clean.

Step 5 is what a chat assistant without tools cannot do. Without it, the agent only guesses that its edit worked.

Two limits matter. The checks see only public URLs, so a change on localhost stays invisible until it is deployed. And some fixes live outside your code, such as a DNS record or a CDN rule. A good server marks those so the agent tells you instead of editing files that cannot fix them.

What to look for in an SEO MCP server

Most SEO MCP servers wrap an existing API, and the wrapper decides how safe it is to hand to an agent that acts on its own. Check these four things.

What to check Why it matters What good looks like
Authentication The key can spend your money A revocable per-workspace key sent as a bearer header, never in the URL
Cost limits Agents retry and loop A per-call price ceiling the agent can pass, and a balance check before batches
Idempotency Timeouts cause retries A retry key that returns the original run instead of charging again
Read-only vs actions Some servers can publish or edit Checks that only read public pages, with code changes left to your agent and your review

The last row deserves a closer look. MCP lets a server describe each tool with hints such as readOnlyHint and destructiveHint, but the tools spec tells clients to treat those hints as untrusted unless the server is trusted. So read what each tool does. A server that only fetches and analyzes pages cannot break your site. A server that writes to your CMS or ad account can, and should require your approval per action.

Also look at the output. An agent works best with findings that name the affected URL, the change to make and how to verify it. A score with a paragraph of prose leaves the agent to guess.

How to connect an SEO MCP server to your coding agent

Our server is at https://seogeoaeo.ai/api/mcp. It uses Streamable HTTP and needs an API key, which you create at /api-keys after signing in. Replace YOUR_API_KEY in each snippet.

For Claude Code, run this once in a terminal. It adds the server for your user across every project.

claude mcp add --transport http --scope user seogeoaeo https://seogeoaeo.ai/api/mcp --header "Authorization: Bearer YOUR_API_KEY"

For Cursor, add this to ~/.cursor/mcp.json. Windsurf takes the same block in its mcp_config.json.

{
  "mcpServers": {
    "seogeoaeo": {
      "url": "https://seogeoaeo.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

For Codex, add this to ~/.codex/config.toml and export SEOGEOAEO_API_KEY in your shell, so the key stays out of the file.

[mcp_servers.seogeoaeo]
url = "https://seogeoaeo.ai/api/mcp"
bearer_token_env_var = "SEOGEOAEO_API_KEY"

The agents page also has configs for VS Code and Claude Desktop. Our SEO API post covers the REST side.

Run agentic SEO checks on your own site

Every public tool on our site is an MCP tool named by its slug, plus run-assessment for a whole-site check. One assessment runs our robots.txt and redirect checks once on the site, then indexing, on-page, schema, readability, citability, freshness and author checks on each page you pass, up to three pages. It returns SEO readiness, AEO readiness and AI access scores from 0 to 100 and one fix list in priority order. The default costs 67 credits for one page and 47 for each extra page. Each check is charged on its own and refunded if it fails.

Each run returns improvement tips with the affected URLs, the action, how to verify it, and a fixIn marker for changes made outside your code. Rerun the assessment on the same URL after you deploy and it reports the percent of earlier issues fixed. Calls accept maxCredits to refuse a run above a price and idempotencyKey so a retry returns the original run. Listing tools, checking credits and reading past runs cost nothing. The tools only fetch public http and https URLs and never write to your site. Your agent makes the edits, in your repo, where you can review them. For what a full audit should cover, see our SEO audit guide.

Keep reading