MCP SERVER

The SEO MCP server.

A local MCP server that hands your AI agent the full SEO pipeline as native tool calls. Keyword research, site audit, fact-checked drafting, and native CMS publish, all from inside Claude Code, Claude Desktop, or Cursor.

PACKAGE BUILT · PUBLISHING TO NPM
claude_desktop_config.json
{
  "mcpServers": {
    "theseoagent": {
      "command": "npx",
      "args": ["-y", "theseoagent-mcp"],
      "env": {
        "THESEOAGENT_API_KEY": "sk_live_..."
      }
    }
  }
}
Exposed tools7 tools
list_projectslist_articlesget_articlegenerate_articlekeyword_researchrun_auditpublish_article
Local stdio server · publishing to npm
BY THE SEO AGENT TEAMUPDATED 2026-06-145 MIN READ
TL;DR

What is the SEO MCP server?

It is a local stdio Model Context Protocol server that exposes The SEO Agent pipeline as seven native tools for AI agents: list_projects, list_articles, get_article, keyword_research, run_audit, generate_article, and publish_article. Add it to Claude Code, Claude Desktop, or Cursor and your agent can research keywords against live data, audit a site, draft a fact-checked article through the quality gate, and publish it to your CMS, all as tool calls. The package is built and publishing to npm; the trial provisions the workspace key it needs. $99 a month, flat, with the REST API included.

7 toolsEXPOSED OVER MCPList, audit, research, generate, publish
LocalSTDIO TRANSPORTRuns on your machine, your key
$99/moFLAT, ALL INCLUDEDMCP, REST, and CMS publishing
WHAT AGENTS HIT TODAY

Your agent can write. It cannot ship SEO.

An AI agent in Claude Code or Cursor can draft a blog post in seconds. What it cannot do is reach real keyword data, audit a live site, or push the finished article to your CMS. Those steps live behind tools it does not have.

  • No real search volume or difficulty, only guesses
  • No site audit, so no idea what is actually broken
  • No fact-check pass, so hallucinated stats ship
  • No native publish, so every article is a copy-paste
  • No memory of what already ranks on your domain
WHAT THE MCP SERVER ADDS

The full pipeline as native tool calls.

The SEO MCP server hands your agent seven tools that run the real pipeline. It lists your projects and articles, researches keywords against live data, audits a domain, generates a fact-checked draft, and publishes to your CMS. Your agent orchestrates; we do the work.

  • keyword_research returns live volume, difficulty, and intent
  • run_audit fetches a domain and reports real findings
  • generate_article runs the gated drafting pipeline
  • publish_article ships natively to your connected CMS
  • list_projects and list_articles give the agent context
WHY THIS MCP SERVER WINS

Four reasons agents ship real SEO through it.

SEVEN TOOLS, ONE PIPELINE

Not a thin wrapper. The same engine the app runs.

The server exposes seven tools that map to real pipeline stages, not toy endpoints. keyword_research hits the same live keyword data feed our product uses. run_audit fetches a domain and returns the same findings the public audit page surfaces. generate_article runs the gated drafting pipeline described on the SEO automation page, and publish_article ships to your connected CMS. Your agent gets the whole loop, not a draft generator.

LIST · GET · RESEARCH · AUDIT · GENERATE · PUBLISH
TOOL MANIFEST · 16:9
FACT-CHECKED, NOT HALLUCINATED

The drafting tool refuses to ship made-up stats.

When your agent calls generate_article, the draft runs through the same fact-check and quality gate the app enforces. Every defensible claim is cited or rewritten, and a draft below the quality threshold is sent back for regeneration rather than returned to your agent as finished. The architecture is the same one we document on the fact-checking feature page. Your agent cannot accidentally publish a hallucination through this tool.

GATE ENFORCED ON EVERY GENERATE CALL
FACT-CHECK GATE · 16:9
LOCAL STDIO, YOUR KEY

Runs on your machine. Your key never leaves it.

This is a local stdio MCP server, not a hosted endpoint your agent calls blind. It runs as a subprocess of your client, reads the workspace key from your local config, and talks to our API directly. There is no third-party MCP relay in the middle. The same key works against the public REST API if you would rather script the calls yourself, and the full reference lives in the API docs.

STDIO TRANSPORT · NO HOSTED RELAY
LOCAL SUBPROCESS · 16:9
PUBLISHES WHERE YOU ALREADY ARE

Drafts land on your live CMS, not a clipboard.

Connect a CMS once and publish_article pushes the finished, formatted article straight to it: featured image, tags, meta, and schema included. WordPress runs through our maintained plugin (see the WordPress connector), and the rest of the connectors are listed on the integrations hub. Your agent closes the loop from keyword to live URL without you touching the editor.

NATIVE PUBLISH · IMAGE · META · SCHEMA
PUBLISH TO CMS · 16:9
HOW TO CONNECT IT

Four steps from key to first tool call.

01GRAB A WORKSPACE KEY

Start the trial and copy your API key from the dashboard. The MCP server authenticates with the same key the public REST API uses, so one key unlocks both surfaces.

02ADD THE SERVER TO YOUR CLIENT

Drop the config block on the right into claude_desktop_config.json (Claude Desktop), your Cursor MCP settings, or your Claude Code project config. The server runs locally over stdio, so your key never leaves your machine.

03CALL THE TOOLS IN PLAIN ENGLISH

Ask your agent to research a keyword, audit a competitor, or draft and publish an article. It picks the right tool, fills the arguments, and streams the result back into the conversation.

04REVIEW WHAT SHIPPED

Every generate and publish call is logged in the app with cost, gate score, and the live URL. The agent does the orchestration; you keep the audit trail and the one-click cancel.

VS A BARE WEBHOOK CONNECTOR

The honest one-screen comparison.

FEATURETHE AGENTEVERYONE ELSE
Exposed as native MCP tools
Keyword dataLive volume, difficulty, intentA bare webhook, no data
Site audit as a tool call
Fact-check + quality gate on drafts
PublishingNative CMS publish via the agentManual copy-paste
TransportLocal stdio, your keyHosted endpoint, shared key
QUESTIONS

What builders ask before wiring it up.

Missing something? Ask us directly.

Q.01What is an SEO MCP server?+

It is a small program that speaks the Model Context Protocol so an AI agent like Claude Code, Claude Desktop, or Cursor can call SEO actions as native tools. Instead of describing a workflow in a prompt, the agent invokes tools such as keyword_research or publish_article directly and gets structured results back. The SEO Agent server exposes seven of these tools that run the real pipeline.

Q.02Is the package on npm yet?+

Not at the moment of writing. The package is built and the publish to npm is imminent. This page is the canonical setup reference, so the config block is correct and ready to paste the moment it lands. The way to start now is the trial: it provisions the workspace key the server needs, so you are set up before the package is even live.

Q.03Which clients does it work with?+

Any MCP client. We test against Claude Code, Claude Desktop, and Cursor. The config shape shown here is the Claude Desktop format with an mcpServers map; Cursor and Claude Code use the same fields. Drop the block into your client config, restart the client, and the seven tools appear.

Q.04What are the seven tools?+

list_projects, list_articles, get_article, generate_article, keyword_research, run_audit, and publish_article. Together they cover the full loop: read your workspace, research a keyword, audit a site, generate a fact-checked draft, and publish it to your connected CMS.

Q.05Does my API key leave my machine?+

No. This is a local stdio server. It runs as a subprocess of your MCP client, reads the workspace key from your local config, and calls our API directly. There is no hosted MCP relay sitting between your agent and us, so the key stays on your machine.

Q.06Can the agent publish a bad article by accident?+

No. The generate_article tool runs the same fact-check and quality gate the app enforces. A draft below the quality threshold is regenerated rather than returned as finished, and defensible claims are cited or rewritten. The agent cannot push a hallucinated draft through to your CMS.

Q.07How is this different from a competitor that ships a bare webhook?+

A webhook is a single endpoint with no data behind it: you still have to bring the keywords, the audit, and the publishing yourself. The MCP server hands the agent the whole pipeline as typed tools, each backed by live data and the quality gate. The agent orchestrates real work, not a forwarding call.

Q.08What does it cost to use the MCP server?+

It is included in the flat plan at 99 dollars a month, the same plan that covers the REST API and native CMS publishing. There is no separate MCP fee and no per-call meter. The first three days are a one dollar trial so you can wire it up and watch it run before you commit.

GIVE YOUR AGENT THE WHOLE PIPELINE.

Wire it up. Watch it ship.

Three days for a dollar. Grab the workspace key, paste the config, and let your agent research, draft, and publish from inside your editor. Cancel in one click if it is not pulling its weight.

FREE TRIAL · CANCEL IN ONE CLICK