
What matters first.
- An SEO MCP server lets an AI client such as Claude, Cursor or ChatGPT query an SEO data source directly, through the open Model Context Protocol. Most connect Google data: rankings, keywords, backlinks, Search Console, analytics.
- What AI answers say about your brand is a different question with a different source: measured answers from ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. FameScan serves them over MCP, with intervals and verdicts.
- A server changes data only through its write tools, within the access you connect it with. In FameScan, reading and changing are separate access levels, “Make changes” starts unticked, and no connection gets more rights than your workspace role.
An SEO MCP server is a connector that lets an AI client such as Claude, Cursor or ChatGPT query an SEO data source directly, through the open Model Context Protocol. Most of them bring Google data into the chat: rankings, keywords, backlinks, Search Console. The question your boss asks now, what ChatGPT says about your brand, needs a different source: measured AI answers.
FameScan is that source as an MCP server: it asks ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews your buyer questions, stores every answer and any sources it cites, and lets Claude, Claude Code, Cursor, VS Code or another MCP client with OAuth read the results, the headline KPIs with a 95% interval and a “real change versus noise” verdict. Facts about other servers come from their official pages, opened on November 12, 2026.
What is an SEO MCP server?
An SEO MCP server sits between your AI client and an SEO tool: it describes what the tool can do, and the client calls it whenever your question needs data. Three parts are involved. The client is the app you type into, such as Claude, Cursor or ChatGPT. The server is the connector a vendor runs. The data source is what sits behind it: a keyword index, your Search Console property, or stored AI answers.
The protocol itself is, in the words of its official site, “an open-source standard for connecting AI applications to external systems.” The site compares it to a USB-C port for AI applications and lists Claude, ChatGPT, Visual Studio Code and Cursor among the clients that support it. According to its governance page, the project is organized under LF Projects, LLC, with a formal governance model and transparent decision-making.
In daily work, an MCP server replaces the detour through export, spreadsheet and paste. You ask, the client fetches the current numbers and writes the answer around them. It does not replace the measurement: what appears in the chat is only as reliable as the tool that collected the numbers.
One note on the words: “MCP SEO” sometimes also means making your own MCP server easy for AI assistants to find. This guide is about the other meaning, using MCP for SEO data and AI visibility data inside your chat.
The question Google data cannot answer: what AI answers say about you
Most roundups of SEO MCP servers sort servers by the Google data they return: rankings, keyword volumes, backlinks, Search Console clicks, analytics sessions. That data answers questions about Google.
The question marketing leads hear now is a different one. When a buyer asks ChatGPT or Perplexity for the best option in your category, are you named, recommended, cited as a source, and who appears instead? No keyword index and no Search Console property holds that answer. It lives in the AI answers themselves, so the source has to be made of AI answers: your buyer questions, asked in each engine, stored, analyzed and counted.
Google data answers questions about Google; a question about what ChatGPT, Perplexity or Gemini say about your brand needs a source made of measured AI answers.
Asking the chat itself is not that source. Type “what do you think of our brand?” into Claude and you get one answer, from one engine, at one moment, to a question with your name already in it. One answer is a snapshot. The pattern is the truth: the same questions without your brand name, across engines, countries and weeks, with enough answers to tell a real change from noise. Our guide to measuring AI visibility explains why a sample beats a screenshot.
Whichever server you connect for this question, ask what its numbers are made of. Which questions, which engines, how many answers, and can you open the answer behind a figure?
Which question needs which SEO MCP server?
Start from the question you are asked most, not from a list of servers: each question has one data source that holds the answer, and the server is only the way in. The table is ordered by question. The examples show what each vendor says on its own official page, as of November 12, 2026; prices and access rules change, so check the linked page before you connect.
| Question and data source | Examples from official pages, November 12, 2026 |
|---|---|
| Where do we rank, for which keywords, and who links to us? The keyword, ranking and backlink index of an SEO suite | Semrush MCP: the SEO API, Trends API data by subscription, read methods of the Projects API. Ahrefs MCP: access to the Ahrefs API on paid plans from Lite; calls use API units. Sign-in: Semrush: OAuth by default, an API key for clients without OAuth. Ahrefs: an approved OAuth connection gets its own MCP-scoped key. |
| Which pages earn clicks from Google Search? Your Search Console property | Community-built servers on top of the Search Console API. Sign-in: your Google account. Check who maintains the server first. |
| What do visitors do on the site? Your analytics property | Google Analytics MCP server from Google: read requests only, through the Admin API and Data API v1. Sign-in: Google credentials with the read-only Analytics scope. |
| What do ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews say about us and our competitors? Stored AI answers to your buyer questions, measured over time | The FameScan MCP server: 17 read tools, 7 write tools. Sign-in: a claude.ai connector with your FameScan sign-in (OAuth), or a workspace API key. Reads by default; writes only if you tick “Make changes” or create a “Read & write” key. |
Read the table by row. The first three rows show how Google sees you and what visitors do. The last one shows what AI engines answer when a buyer asks. Connect one server per row you are really asked about, and nothing more. That is also the honest answer to “which marketing MCP should we add?”: the one whose data answers your next question.
Can an SEO MCP server change your data?
Only if it has tools that write and the access you connected it with allows them. A server offers a list of tools; each one either reads or changes something, and what your client may call depends on the access you approved when you connected.
That turns the worry behind “MCP is bad for SEO” into two questions you can check. Can the server change anything? Its access levels tell you: Google, for example, says its Analytics server serves read requests only. Are the numbers right? That depends on the source: your chat client writes the sentences, but the figures should come straight from the tool, with enough detail to verify them. Two checks help: ask which call to the server each figure came from, and ask only for what that source holds. Otherwise the model can fill in a number the source does not have from its own memory.
In FameScan, reading and changing are two separate access levels. Reading covers 17 tools: projects, visibility KPIs, the trend, competitors, the per-engine breakdown, results per prompt, stored answers, cited sources, source gaps, items from the Action Center, runs and credits, alerts, the AI readiness audit and the methodology. Changing covers 7: adding prompts, pausing or activating them, adding or removing a competitor, updating an action, starting a measurement run and running an AI readiness audit.
When you connect through sign-in, the consent screen shows “Read AI-visibility data” as a fixed tick and “Make changes” as a separate box that starts empty. A connection approved for reading is never even offered the tools that change something. Access is capped at your role in the workspace: someone who can only read can only grant reading. An API key acts as the person who created it, never with more rights than its scope.
Reading spends no Answer Credits; starting a run does. The visibility overview returns the headline KPIs with their 95% interval and the number of answers behind them, and marks changes against the previous period as real change, noise, or too few answers to tell. Every stored answer can be opened by its ID. All connections are listed under Integrations, Connected apps, where you can disconnect any of them at any time.
Is an MCP an API?
Not quite: an API is called by code you write, while an MCP server is called by your AI client in the middle of a conversation. The server describes each tool in plain words, with its inputs, and the client decides which one your question needs.
Many MCP servers sit on top of an API, and the difference decides what you build. For a question you ask once, the chat is faster. For a report that has to run every Monday without anyone asking, a fixed API call in a workflow is steadier.
FameScan offers both over the same tools: the MCP server for Claude, Cursor and other clients, and a REST API for n8n, Make, Zapier or scripts, with a workspace API key. Both read the same numbers as the dashboard and the Copilot in the portal, so a figure in Claude matches the portal for the same period and filters.
How do you connect FameScan to Claude, Claude Code or Cursor?
Everything starts on the Integrations page of the FameScan portal, which shows the server address and a ready snippet to copy for each client.
- claude.ai. Open Settings → Connectors → Add custom connector, paste the URL and click Connect. You sign in to FameScan and pick the workspace and the access; no API key needed. Other MCP clients that support OAuth connect the same way, which makes this the shortest route to a Claude SEO MCP setup.
- Claude Code. An admin or owner creates an API key under Integrations. It defaults to “Read only”; choose “Read & write” if Claude Code should also add prompts or start runs. Run the command from the snippet in your terminal, then type /mcp to check the connection.
- Cursor. The JSON snippet goes into .cursor/mcp.json in your project or into Cursor’s global MCP settings. Snippets for Claude Desktop and VS Code sit right next to it.
A new key is shown once, and the snippets already contain it. Any member can connect an app for themselves through sign-in; API keys are managed by admins and owners. The same page lists four ready-made workflows as MCP prompts: a weekly AI-visibility pulse, a competitor radar, “Win a prompt” and an AI-search content brief.
The MCP server and the REST API come with every paid plan, and a running trial counts. Lite costs $12 a month, or $9 a month billed yearly, with 20 Answer Credits a month. Starter costs $49 a month, or $39 a month billed yearly, with 200 Answer Credits and daily runs, checked November 12, 2026. Every plan is on the pricing page, and setup questions are answered on the help page.
If nobody on your team works in an AI client, the Copilot in the portal answers from the same tools, with charts; there, every change arrives as a proposal you apply.
Worked example: the weekly pulse in one chat
On Monday the analyst picks the workflow “Weekly AI-visibility pulse” in Claude. It has Claude call the visibility overview for the last 7 days, the competitor comparison, the Action Center and the weakest prompts. The reply keeps what counts and leaves out the noise.
In this example it comes back like this. The FameScan score is 31, with an interval of 24 to 38. The one real change: the recommendation rate is up 9 points. Paylark moved the most. Among the three weakest questions is “best invoicing app for freelancers”. At the top of the Action Center sits “Get present on sources AIs cite for competitors”.
For the weakest question she starts “Win a prompt”. Claude pulls the latest answers per engine, who was named in which order and which domains are cited most, then opens two stored answers. In both, the engine leans on the same two comparison pages, where Paylark is listed and Quillbase is not.
In the same chat she asks the analytics server for sessions on the pricing page. The two stay apart: one row says what visitors did, the other what AI answers said.
That is the Monday test. She shows the client the rate with its interval, the verdict, and the two stored answers with their sources, and every number leads to an answer that opens in the portal. With a “Read & write” key, the same chat could add new buyer questions from the plan and move the action to in progress; here she does that in the portal.
Which MCP servers do you actually need for SEO work?
As few as the questions you really have to answer. Every server is one more access someone maintains and approves, and a chat with five sources does not get smarter, only longer.
- In-house SEO lead: one server for your own site, Search Console or analytics, and one source for AI answers. Add the research suite if you already use it every day.
- Agency: one suite server for research across clients, and one FameScan workspace per client, because every connection belongs to exactly one workspace. Client numbers stay apart. More on the page for agencies.
- Founder doing marketing: start with the question investors or customers ask. Often that is the one about AI answers, with analytics as the second source.
Connect the one server that holds the answer to the question you hear most, and add the next only when a new question needs it. Once FameScan is connected, paste this into your chat as your first prompt:
The reply tells you where your next hour goes. If your buyer questions are still thin, the work starts with the questions themselves: our guide to prompt tracking shows how to build a question set that reflects real buyers.
Frequently asked
What is an SEO MCP server?
A connector that lets an AI client such as Claude, Cursor or ChatGPT query an SEO data source directly, through the open Model Context Protocol. Instead of exporting tables and pasting them into a chat, you ask a question and the client fetches the current numbers from the connected tool. The numbers are only as good as the tool that measured them.
Is an MCP an API?
Not quite. An API is called by code you write, with fixed calls. An MCP server describes its tools so an AI client can pick and call them in the middle of a conversation. Many MCP servers sit on top of an API. FameScan offers both over the same tools: the MCP server for Claude, Cursor and other clients, and a REST API for n8n, Make, Zapier or scripts.
Which MCP servers do SEOs use?
Mostly the servers of tools they already pay for: a keyword and backlink suite for research, plus Search Console and Google Analytics for their own site. For the question of what ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews say about a brand, they need a source made of measured AI answers, such as the FameScan MCP server.
Can an MCP server change my data?
Only through tools that write, and only within the access you connect it with. Check which tools a server lists and which access you approve. In FameScan, reading and changing are separate: at sign-in, “Make changes” is unticked, a read-only connection is never offered the tools that change anything, and no connection gets more rights than your workspace role. You can disconnect any app under Integrations at any time.
Do I need a paid plan to use the FameScan MCP server?
Yes. The MCP server and the REST API come with every paid plan, and a running trial counts. Lite costs $12 a month, or $9 a month billed yearly, with 20 Answer Credits a month, checked November 12, 2026. The other plans are on the pricing page.
Can I just ask ChatGPT what it says about my brand?
You will get one answer, from one engine, at one moment, to a question that already contains your brand name. That is a snapshot. It becomes evidence only across many answers to real buyer questions without your name, in several engines and over weeks, with an interval that tells a real change from noise.
Sources & further reading
Sources checked on 12 November 2026.