Part of SEO Agents in 2026: What an AI SEO Agent Can Actually Do
The Best MCP Servers for SEO in 2026
The best SEO MCP servers for Search Console, indexing, crawling, rank and backlink data, compared by what they read, what they change and what they cost.
An SEO MCP server connects an AI client such as Claude, Cursor or ChatGPT to a search tool or data source, so the model can query real data or take an action instead of guessing. The best MCP servers for SEO fall into five groups: Search Console and analytics, indexing, crawling and audits, rank and backlink data, and servers that edit your site. Which ones you need depends on what the agent should read, what it may change, and what each call costs.
This list compares the main servers on those points, as of October 2026. The SEO agent guide covers what an agent built on them can do and the guardrails it needs. One rule holds for every server below: none can make Google crawl, index or rank a page.
SEO MCP servers at a glance
| Server | Reads | Acts | Auth and cost |
|---|---|---|---|
| Search Console servers (mcp-gsc) | Search Analytics, URL Inspection, sitemaps | Sitemap submissions; deleting sitemaps and adding or deleting sites only behind a flag | Your Google OAuth client or a service account; Google's API quotas |
| Google Analytics MCP | GA4 reports, funnels, realtime | One annotation tool, off by default | Google Application Default Credentials; open source |
| IndexCyborg | Stored URL Inspection results, technical checks, reports | IndexNow notifications, sitemap rescans, a Request indexing queue | Scoped API key; subscription or page packs |
| Bing Webmaster Tools (community) | Bing search performance, crawl issues | URL submission to Bing and IndexNow | Bing Webmaster API key or OAuth |
| Screaming Frog SEO Spider | Crawl data, reports, bulk exports, page content | Starts crawls, writes files; Node.js scripts once enabled | Runs locally; paid licence |
| Chrome DevTools MCP | Lighthouse audits, performance traces, network, rendered page | Drives a real browser: clicks, forms, scripts | Runs locally; open source |
| Firecrawl | Scraped pages, site URL maps, crawls, web search | Crawl jobs and page monitors in your account | OAuth or API key; credits |
| Ahrefs | Site Explorer, Keywords Explorer, Rank Tracker, Site Audit, GSC data | An account management group (projects, keyword lists) | OAuth or MCP key; API units, Lite plan and up |
| Semrush | SEO and Trends API data, Projects (read-only) | None documented | OAuth 2.1 or API key; API units |
| DataForSEO | Any DataForSEO endpoint: SERP, keywords, backlinks, on-page | Every call is a paid API request | API login or OAuth; pay as you go |
| SE Ranking | Keywords, backlinks, rank tracking, audits, AI visibility | Not documented per endpoint | OAuth 2.1 or an account API key; Data API credits |
| WordPress MCP Adapter, Webflow | Your content and pages | Exposed abilities (WordPress); meta tags, CMS items, publishing (Webflow) | Your site's own permissions |
“Acts” means the tool changes something outside the conversation: a submission, a crawl that spends credits, an edit to a live page.
How to compare MCP SEO tools
- Whose numbers it reads. Search Console and Analytics return Google's own data for your properties. Ahrefs, Semrush, DataForSEO and SE Ranking return estimates from their own databases. Crawlers return what they fetched.
- What it can change. A server can label tools with annotations such as
readOnlyHintanddestructiveHint, but the MCP schema calls them hints and says clients should never base tool use decisions on annotations from untrusted servers. Read the tool list yourself. The tools specification also says a human should always be able to deny a tool call. - How it signs in. A local server runs on your machine with your credentials, usually over stdio. A remote server is a URL called over Streamable HTTP, with an OAuth sign-in or an API key in a header.
- What a call costs. Google's APIs are bounded by quotas. Data providers charge per call, in API units or credits, so an agent that loops spends your plan.
Search Console and analytics servers
Google Search Console servers
Google doesn't offer a managed Search Console MCP server: as of October 2026, neither Search Console nor Google Analytics is on Google's list of Google-managed MCP servers. Every Search Console server wraps the public Search Console API, which has four resources (Search Analytics, Sitemaps, Sites and URL Inspection) and no method that requests indexing.
mcp-gsc is a widely used open-source option: MIT-licensed, about 2,100 GitHub stars, started with uvx mcp-search-console. It signs in with your own OAuth desktop client or a service account, reads performance data, inspects URLs in batches of up to 10, and lists and manages sitemaps. The tools that add or delete a site, or delete a sitemap, stay off unless you set GSC_ALLOW_DESTRUCTIVE to true.
Google's usage limits apply to every server: 2,000 URL inspections a day and 600 a minute per site, and 1,200 Search Analytics queries a minute per site and per user. Submitting a sitemap needs the full webmasters scope (sitemaps.submit). The Google Search Console MCP guide compares hosted alternatives and gives a config for each client.
Google Analytics MCP server
The googleanalytics organization on GitHub publishes an Analytics MCP server, labeled experimental, Apache-2.0, started with pipx run analytics-mcp. It signs in with Application Default Credentials and the analytics.readonly scope, and runs GA4 reports, funnel reports and realtime reports. Its only write tool, create_property_annotation, needs the --allow-write flag and the analytics.edit scope.
Indexing servers
Lists of SEO MCP servers rarely cover this group, and it is where reading and acting differ most. An agent has three documented ways to act on indexing: submit a sitemap through the Search Console API, notify the IndexNow engines, and use Request indexing, which exists only as a button in the URL Inspection tool, with a daily limit Google doesn't publish. The Indexing API is not a fourth route for ordinary pages: Google says it can only be used for pages with JobPosting or BroadcastEvent embedded in a VideoObject (Indexing API quickstart).
IndexCyborg
IndexCyborg's server is remote, at https://indexcyborg.com/api/mcp, with a scoped API key. Its read tools return Google's stored URL Inspection result for each inspected page in your sitemaps, plus technical checks and reports. Its act tools rescan sitemaps (discover_urls), notify IndexNow engines of up to 1,000 URLs per call (notify_search_engines), and queue URLs for Request indexing (request_google_indexing). It doesn't read Search Analytics, so it runs next to a Search Console server. Details in the last section.
Bing Webmaster Tools servers
The Bing Webmaster Tools API takes OAuth 2.0 or a per-user API key that works for all of that user's verified sites, and the Bing servers in MCP directories are community projects. bing-webmaster-mcp, for example (MIT, one maintainer), reads search performance and crawl issues, and its submit_urls tool sends URLs to both the Bing Webmaster API and IndexNow. Its description promises immediate indexing; the IndexNow FAQ says a submission doesn't guarantee it.
As of October 2026, the IndexNow engines are Bing, Yandex, Seznam.cz, Naver, Yep, the Internet Archive and Amazonbot, not Google, and a 200 only means the URL was submitted successfully (documentation). See IndexNow MCP and request Google indexing from an AI agent.
Crawling and technical audit servers
Screaming Frog SEO Spider
Since version 24, the SEO Spider can run as an MCP server. It is a paid licence feature, at £199 per licence per year as of October 2026 (pricing). It runs headless over stdio from a Claude Desktop extension, or as a Streamable HTTP server started from the app's MCP menu. Its tools start crawls, generate reports and bulk exports, and return a page's content, links and screenshot. Its file tools act on your machine: sf_write_text_file overwrites existing files without warning, and the Node.js tools are off by default. The guide describes no authentication, so keep it local.
Chrome DevTools MCP
The server from the ChromeDevTools organization on GitHub (Apache-2.0) starts with npx -y chrome-devtools-mcp@latest and drives a real Chrome. Per its tool reference, lighthouse_audit returns Lighthouse scores and failing audits for accessibility, SEO and best practices (performance is excluded), performance_start_trace records a trace for Core Web Vitals, and emulate sets the user agent, network, CPU and viewport. It can also click, fill forms and run scripts, so it acts on whatever page it opens. By default it may send trace URLs to Google's CrUX API and collects usage statistics; --no-performance-crux and --no-usage-statistics turn both off.
Firecrawl
Firecrawl's server (MIT) scrapes pages to Markdown or JSON, maps a site's URLs and runs crawls, locally with a FIRECRAWL_API_KEY or remotely with a sign-in or bearer key. Most calls spend credits. It shows a page as Firecrawl fetched it; Google's view comes from URL Inspection.
Rank, keyword and backlink data servers
These are the servers most SEO MCP lists are about. Keyword volumes, rankings and backlink counts come from each provider's own database, so they are that provider's estimates, not Google's figures. All four meter usage, in API units, account balance or credits.
Ahrefs
Ahrefs runs a remote server only, at https://api.ahrefs.com/mcp/mcp (Ahrefs MCP docs); local setup is no longer supported. You connect with OAuth, which creates a dedicated key, or with an MCP key you generate, on paid plans from Lite up. A tools parameter in the URL picks tool groups such as keywords-explorer, site-explorer-reports, rank-tracker, site-audit and gsc, or a curated essentials set for clients that cap the number of tools. A management group covers projects and keyword lists, so read those tools before you allow them.
Calls spend API units shared with the Ahrefs API. As of October 2026, the help center lists 200,000 units a month on Lite, 800,000 on Standard and 2,000,000 on Advanced, at least 50 units per call, and 100, 250 or 500 rows per request. On Lite, that is at most 4,000 calls a month. Ahrefs also says the endpoint is meant for interactive use in MCP clients: custom scripts and bridges aren't permitted, and code should use its public API.
Semrush
Semrush's server is at https://mcp.semrush.com/v2/mcp, Streamable HTTP only, with OAuth 2.1 by default and an Authorization: Apikey header for agents without OAuth (Semrush MCP docs). It exposes the SEO API and Trends API endpoints your subscription covers, and the Projects API in read-only methods: it can't create projects or change settings. As of October 2026, Semrush One Starter, Semrush One Pro+, SEO Classic Pro and SEO Classic Guru include 50,000 API units that refresh on renewal; SEO Classic Business and Semrush One Advanced need a units package. Each request spends units according to the report and the amount of data returned.
DataForSEO
DataForSEO's server (Apache-2.0) is deliberately small: three tools browse and search DataForSEO's documentation, and one, api_request, calls any endpoint. That reaches all of its APIs, including SERP, Keyword Data, Backlinks, On-Page and DataForSEO Labs. Locally it runs with npx dataforseo-mcp-server and your API credentials (not your dashboard login) in DATAFORSEO_LOGIN and DATAFORSEO_PASSWORD; over HTTP it defaults to OAuth, and DataForSEO also hosts an endpoint. As of October 2026, pricing is pay as you go, with a $50 minimum top-up and a $1 trial credit. Every call spends balance, so cap how many calls an agent makes in one run.
SE Ranking
SE Ranking hosts its server at https://api.seranking.com/mcp over Streamable HTTP, with OAuth 2.1 or an API key in an X-Api-Key header. It spans the Data API (keywords, backlinks, domains, SERPs), the Project API (projects, rank tracking, audits, competitors), local marketing data and SE Visible's AI visibility data. Two caveats from its page: a key has the same access as your account, because restricted or read-only keys aren't available yet, and not every API endpoint is available through MCP. Per its credits system, every successful Data API call spends credits, while Project API calls count against your subscription limits.
Servers that change your site
The last group acts on the site itself. The WordPress MCP Adapter, the official WordPress package for MCP, exposes abilities registered through WordPress's Abilities API as MCP tools, over HTTP or WP-CLI. Abilities stay private until they are marked public. Webflow's MCP server can update meta titles, descriptions and Open Graph tags across pages, bulk-update CMS items and publish them. An agent there can do only what you can do in the Designer, and its changes land in the site's activity log.
Keep these behind approval. A title rewrite across a whole site is one tool call, and so is a mistake.
Which SEO MCP servers to combine
No single MCP server for SEO covers every job, and clients load several at once. Connect the ones a task needs: fewer tools are easier for a model to choose from.
- Performance questions: a Search Console server, plus Analytics for what happens after the click.
- Keeping new pages indexed after each deploy: an indexing server next to Search Console.
- Technical audits: Screaming Frog for a site-wide crawl, Chrome DevTools for one page.
- Keyword and backlink research: one data server, whichever plan or balance you already have.
- Edits: a CMS server, or a coding agent in your repository, with every write approved.
For servers outside search, such as code, databases and deploys, see the best MCP servers for developers.
A Claude MCP setup for SEO
In Claude Code, three commands connect an indexing server, a keyword data server and a browser (syntax from the Claude Code docs; the Semrush line is Semrush's own):
claude mcp add --transport http indexcyborg https://indexcyborg.com/api/mcp \
--header "Authorization: Bearer ic_your_key"
claude mcp add semrush https://mcp.semrush.com/v2/mcp -t http
claude mcp add --transport stdio chrome-devtools -- npx -y chrome-devtools-mcp@latestSemrush uses OAuth by default: run /mcp in Claude Code, select the Semrush server and choose Authenticate, which opens Semrush's login page. After that, one prompt can draw on all three:
Which /blog pages does Google report as not indexed? Run a Lighthouse
audit on two of them, and list the keywords each one ranks for in Semrush.Claude Code MCP covers scopes and project config, and MCP vs API covers when a script should call a REST API instead.
Before you connect an SEO MCP server
- Trust the server. Claude Code's docs warn that servers that fetch external content can expose you to prompt injection. Crawlers and scrapers read pages other people wrote.
- Narrow the credential and cap spending. Prefer read-only or scoped keys where a provider offers them, and set unit limits where they exist: keys that Ahrefs creates through OAuth start with a monthly limit of 200,000 API units, which workspace admins can change or remove (Ahrefs MCP docs).
- Approve every act tool. Let read tools run and require a prompt for submissions, crawls and edits. The SEO agent guide linked above has a permission config for Claude Code.
- Read results in the engine's words. “Submitted”, “accepted” and “requested” are not “indexed”. Google's recrawl guide says crawling can take from a few days to a few weeks.
How IndexCyborg fits in an SEO MCP stack
IndexCyborg is the indexing server in this list: remote and stateless, with API keys scoped to read, notify and manage, optionally limited to some websites. It inspects the URLs in your sitemaps with the URL Inspection API, new pages straight away within Google's 2,000 inspections per property per day, re-checks them after 7 days, and stores coverageState word for word. get_unindexed_pages never lists uninspected pages.
The gates sit on the server, where a prompt can't remove them: no IndexNow notification before the domain and key file are verified, and Request indexing only through the IndexCyborg Chrome extension (loaded from the dashboard's Settings), which clicks the button in your own Search Console session at most 10 times per property per rolling 24 hours. A requested URL is queued for crawling, not indexed. For keywords, rankings, backlinks or Search Analytics, pair it with the servers above. Starter is $5 a month for 1 website, or page packs start at $5 (pricing). Setup is in check indexing with MCP.
Frequently asked questions
Is there a free SEO MCP server?
Yes. Open-source Search Console servers such as mcp-gsc are free to run with your own Google credentials, within Google's API quotas. Chrome DevTools MCP and the Google Analytics MCP server are open source as well. Keyword, rank and backlink servers need a paid plan or a prepaid balance, and Screaming Frog's MCP server needs a paid SEO Spider licence.
Do the Ahrefs and Semrush MCP servers cost extra?
Ahrefs says there is no extra fee for its MCP server, but every call spends API units from your plan, from Lite up. Semrush includes MCP access with 50,000 API units on several plans and needs a units package on others. An agent that loops can use up an allowance quickly.
Can an MCP server get my pages indexed on Google?
No server can make Google index a page. A server can submit a sitemap, notify IndexNow engines such as Bing, or queue a URL for Request indexing in Search Console, and each of those only puts a URL in front of a crawler. Google says requesting a crawl doesn't guarantee that the page will be included in search results at all.
Does Google have an official Search Console MCP server?
Not as of October 2026. Search Console isn't on Google's list of Google-managed MCP servers, so every Search Console MCP server is a third-party wrapper around the public Search Console API. Google Analytics has an experimental open-source server from the googleanalytics organization on GitHub.
Is it safe to give an AI agent write access through MCP?
Only with limits the agent can't override: credentials scoped to what it needs, caps enforced by the server, and a confirmation prompt for every tool that submits, spends or edits something. Start with read tools and add write tools once you trust the loop.