Best SEO MCP servers: 11 checked, with prices and setup

Search for the best SEO MCP servers and the top results are mostly written by vendors who put their own server first: Nightwatch's list opens with Nightwatch, SE Ranking's with SE Ranking, Serpstat's with Serpstat. This list is written by a vendor too. squirrelscan is one of the eleven servers below, and we make it.

An SEO MCP server is a Model Context Protocol server that gives an AI agent SEO tools it can call: keyword volumes, backlinks, rankings, your Search Console numbers, a crawl of your site. The agent fetches the data itself, in the session where it is already editing your code, instead of you pasting a report into a chat window.

The servers that get called SEO MCP servers do three different jobs. Most sell keyword, SERP, backlink and rank data from the vendor's own index. A few read your own Google data. A few look at your pages and report what is broken. Pick by job first; the price comparison only makes sense inside a job.

How we checked

Every endpoint, sign-in method, price and scope below comes from the vendor's own docs or pricing page, read on 23 October 2026, and links there. Where a vendor does not publish something, or two of its own pages disagree, we say so. squirrelscan's rows come from our code, not our marketing.

Every server compared

Prices read 23 October 2026 on each vendor's own site, monthly billing unless the row says otherwise. Nightwatch showed us euros.

Keyword, backlink and rank data

These are the servers most roundups mean. Each one puts a vendor's existing API behind MCP tools, so the data is the same data you would get from that vendor's dashboard or REST API, and so is the bill. If you already pay for one of these suites, its server is the one to connect.

Semrush MCP

Semrush MCP is hosted at https://mcp.semrush.com/v2/mcp, with OAuth by default and an Authorization: Apikey header for clients that cannot do OAuth. It serves Semrush API data: keyword and domain reports, backlinks, and your projects, which it can read but not create or change. Traffic and market data needs a separate Trends API plan. Every call spends API units. Semrush One Starter, $199 a month or $165.17 billed annually, includes 50,000 units that refresh on renewal; SEO Classic Business and Semrush One Advanced need a unit package on top.

Ahrefs MCP

Ahrefs MCP is hosted at https://api.ahrefs.com/mcp/mcp for accounts on Lite or above, which is $129 a month. Calls spend API units from the plan's monthly allowance, and a key created through the OAuth flow is capped at 200,000 units a month by default. Ahrefs limits the endpoint to AI clients: calling it from your own scripts or bridges is not permitted, so code you write should use their REST API.

SE Ranking MCP

SE Ranking MCP is hosted at https://api.seranking.com/mcp, with OAuth 2.1 or an API key. It reaches four SE Ranking products: the Data API (keywords, backlinks, domains, SERPs), your projects (rank tracking, audits, competitors), local marketing data and SE Visible's AI visibility data, subject to your subscription. API access comes with every SE Ranking plan, from Core at $129 a month, and pay-as-you-go API credit starts at $50 for 250,000 credits. One caution: restricted and read-only keys are not available yet, so a key carries your account's full access.

DataForSEO MCP

DataForSEO's server runs hosted at https://mcp.dataforseo.com/v3/mcp, or locally with npx dataforseo-mcp-server or their Docker image, and signs in with your API login and password. It wraps their APIs: SERP, keyword data, DataForSEO Labs, backlinks, on-page, AI optimization, business data and domain analytics. There is no subscription. Billing is pay as you go with a $50 minimum top-up, balances do not expire, and new accounts get $1 of trial credit. Several other tools resell this data, ContextBolt among them.

Nightwatch SEO MCP

Nightwatch is a rank tracker, and its server exposes your rankings, AI visibility, keyword research, site audit findings and saved reports. It runs as a hosted endpoint or an npx package, with a scoped API key that is read-only unless you opt into write scopes. Plans start at €79 a month billed yearly on its pricing page. Nightwatch's own pages disagree on access: the pricing FAQ says the MCP is on every plan including Starter, while the MCP page says Professional and above, and not during the free trial. Ask them before you buy Starter for the MCP.

ContextBolt SEO

ContextBolt SEO is a hosted server at https://seo.contextbolt.app/mcp, aimed at founders doing their own SEO. It bundles keyword, SERP, backlink and AI visibility data, site audits of up to 50 pages a run and a read-only Search Console connection that spends no credits, for $35 a month with 1,000 credits after a 7-day trial. Its estimates come from DataForSEO, by its own account. It launched on 20 May 2026.

Your own Google data

Search Console: community servers only

Google has not shipped a Search Console MCP server. Its list of official MCP servers and the Google Cloud MCP supported-products page cover Google Analytics, BigQuery, Maps, Workspace and a long list of Cloud products, with no Search Console entry. We checked both on 23 October 2026.

What exists are community servers on the public Search Console API. mcp-gsc, with over 2,000 GitHub stars, is MIT-licensed, runs locally with uvx mcp-search-console, and signs in through your own Google OAuth client or a service account. It covers search analytics, URL inspection, indexing checks and sitemaps. Its changelog shows why maintenance matters: the MCP Python SDK 2.0, released 28 July 2026, broke every fresh install until the project pinned the older SDK. Look at the last commit before you depend on any of them; mcp-gsc's latest push was 15 September 2026.

Google Analytics MCP

Google does publish one for Analytics. The Google Analytics MCP server is open source under Apache 2.0, runs locally on Python 3.10 or later through pipx or uvx, signs in with Application Default Credentials, and is read-only. It lists your accounts and properties and runs standard and realtime reports. It costs nothing beyond a Google Cloud project with the Analytics Admin and Data APIs turned on.

Servers that look at your site

The data servers tell you what to target, and a few of them pass along findings from the vendor's own site audit. The three below are built around the crawl itself: they fetch your pages, or read a crawl you ran, and report what is broken.

Screaming Frog SEO Spider MCP

Screaming Frog added an MCP server in SEO Spider 24.0, released 19 May 2026. Claude, LM Studio and other assistants can run crawls and analyse, export and reshape the crawl data. It runs on your own machine: the assistant launches the Spider headless over STDIO, or you start the MCP server in the app and connect to http://localhost:11435/mcp. It is a paid licence feature that will not work in the free version, so budget £199 per user per year. Read the Node.js setting before you enable it: it lets the model install Node packages and run its own scripts on your machine. Screaming Frog says not every SEO Spider feature works through the MCP yet. For the crawler itself against its alternatives, see Screaming Frog alternatives.

Sitebulb MCP

Sitebulb's MCP connects Claude or ChatGPT to your projects, audits, issues and URL data, from Sitebulb Desktop or Cloud. It is included on every plan and the free trial; Desktop Lite is $18 a month. It is read-only. It reads finished audits and does not start crawls, so the crawl still runs or gets scheduled in Sitebulb, and an agent cannot re-run one to check its own fix.

squirrelscan

Ours. The hosted server at https://mcp.squirrelscan.com/mcp runs the audit itself: it crawls the URL, runs 300+ audit rules across SEO, performance, security, accessibility, content and agent experience (AX), and returns the report. The agent can start another audit after it changes the code, which is how it checks its own fix. Hosted audits cost 50 credits plus 2 per audited page, failed audits are refunded, the free plan includes 500 credits a month, and Pro is $19 a month.

There is also a local stdio server, squirrel mcp, built into the CLI. Signed out, or with offline: true, it audits on your machine and costs nothing.

What it does not do: keyword volumes, backlinks, rankings, or tracking brand mentions in AI answers. For those, connect one of the data servers above next to it.

Which ones to connect

Covering both sides takes two servers: one for data, one that looks at the site.

Setting up squirrelscan's hosted MCP

The endpoint is the same for every client:

https://mcp.squirrelscan.com/mcp

It speaks streamable HTTP. Authenticate with OAuth, a browser sign-in that suits a person at an editor, or with an API key as a bearer header, which suits CI and shared configs.

Claude Code. One command, then the OAuth flow:

claude mcp add --transport http squirrelscan https://mcp.squirrelscan.com/mcp

Then run /mcp inside a session, select squirrelscan, and choose Authenticate.

Cursor. Add this to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "squirrelscan": {
      "url": "https://mcp.squirrelscan.com/mcp"
    }
  }
}

ChatGPT, Codex, VS Code and the other clients are covered in the MCP client setup docs. For a static key instead of a browser sign-in, mint one with squirrel keys create --shell, or at app.squirrelscan.com/settings/api-keys, and send it as an Authorization: Bearer header.

What tools it exposes

The hosted server has 27 tools. The core loop is run_audit to crawl a URL, get_audit_status to poll while it works, and get_report to fetch the result as summary JSON, compact LLM text or full markdown. A run_audit call without confirm returns a credit estimate instead of spending, unless the run is cheap enough to fall under the auto-run threshold. max_pages sizes the crawl for that run, and max_pages: 1 audits just the URL you gave it, the cheapest way to verify one page.

Past the core loop: list_issues and get_issue work a per-website issue tracker, and comment_on_issue lets the agent note what it fixed. list_rules and get_rule explain why a check failed and how to fix it. The entity tools (list_entities, get_entity, get_entity_graph) read the structured data a site declares. Call whoami and get_credit_balance first, since they tell the agent which organization it is acting for.

Two flags on run_audit decide what an audit leaves behind. ephemeral: true returns the report without registering the site, which is what you want for a one-off check: without it, every throwaway check becomes a tracked site and uses a slot under your plan's website limit. kind marks a site you keep as owned or prospect, and list_websites filters on it, so an agent cleaning up a few hundred cold-outreach audits can ask for the prospects only. delete_websites then removes up to 50 at a time and echoes the domain behind every id it touches.

Try it locally first

A signed-out local audit runs the local rules for free, with no account and no MCP client (the cloud-backed checks need a signed-in run):

curl -fsSL https://install.squirrelscan.com | bash
squirrel audit https://example.com

For the full tool reference see the hosted MCP docs, or go straight to setup for Claude Code, Cursor or ChatGPT. To see what a report looks like first, browse a sample audit or point the website audit tool at your own URL.