Google Search Console MCP: Ask an AI Assistant About Your Search Data
A Google Search Console MCP server lets an AI assistant query your search data in plain language. What it can do, how it works, which servers exist and how to set it up safely.
02 / In this article
A Google Search Console MCP server is a small program that lets an AI assistant read your Search Console data through the Search Console API, so you can ask questions in plain language instead of building filters by hand. As of October 2026, we found no server published by Google for Search Console, only community projects, so the setup deserves a careful look before you connect anything. This guide explains what the connection can answer, how it works, which servers exist, which limits still apply and how to set it up with the least risk.
Key Takeaways
- A Search Console MCP server is a translator. The assistant asks a question, the server turns it into a Search Console API request, and the answer comes back as data the assistant can read and explain.
- No official Google server was found. As of October 2026 the servers we checked are community projects. Treat them like any code you run on your own machine: read it first.
- Start read-only. Google offers a read-only API scope. Use it, with a dedicated account, and give write access only when a task truly needs it.
- The data has limits. Search Console keeps 16 months, the API returns top rows rather than guaranteeing every row, and quotas apply. The assistant cannot see what Google does not return.
- The assistant reads, you decide. Use it to find patterns and draft a list of fixes. A person checks the findings before any page changes.
What is a Google Search Console MCP server?
A Google Search Console MCP server is a program that exposes Search Console data to an AI assistant as a set of tools the assistant can call. MCP stands for Model Context Protocol, which its documentation describes as an open-source standard for connecting AI applications to external systems. The same documentation compares it to a USB-C port: one standard way to plug an assistant into a data source.
In practice you install or connect the server, sign in to Google, and tell your assistant that the server exists. When you ask "which pages lost clicks last month?", the assistant picks a tool, sends the parameters and reads what comes back. If you are new to the protocol, our explainer on what MCP is covers the basics, and our overview of MCP tools for marketers places this server next to the others. The assistant is only as good as the data under it, so if your tracking is not trustworthy yet, start with analytics and tracking.
How does it work under the hood?
The server calls the Search Console API on your behalf, using credentials you provide, and returns the result to the assistant. The assistant never touches Google directly. It only sees what the server hands back.
Two details matter for safety and for accuracy.
- The scope sets what the server can do. Google's API documentation lists two scopes for Search Console:
webmasters.readonly, described as read-only access, andwebmasters, described as read/write access. A server that asks for the first one cannot change your account, whatever the assistant tells it to do. - The API method defines the data. The Search Analytics method returns performance rows grouped by dimensions such as query, page, country, device and date. It accepts filters, a row limit and a start row for paging, plus a data state:
finalfor finalized data,allto include fresh data.
Where the server runs depends on the project. Local servers run on your own computer and talk to your assistant through standard input and output. The MCP documentation notes that a local server runs with the same privileges as the client that started it, which is why you read the code first. We come back to that in the safety section.
What can you ask a Search Console MCP server?
You can ask anything that the Search Console data can answer, in plain language: queries, pages, countries, devices, click-through rate, position, sitemaps and the index status of single URLs. The assistant then works out the filters and, if you ask, explains what it found.
Search Console data covers four main areas, and the servers we checked map their tools to them:
- Search performance. Clicks, impressions, CTR and average position, grouped by query, page, country, device or date.
- URL inspection. The index status of a single URL, as Google last saw it.
- Sitemaps. Which sitemaps are submitted and what Google reports about them.
- Properties. Which sites the account can see.
The tool names differ from one server to another. One project we read documents tools such as list_properties, query_performance, top_queries, top_pages, compare_periods and inspect_url. Another documents a single search_analytics tool with a quick-wins mode. Read the README of the server you choose, and check its last update date before relying on it.
Want a measurement setup you can trust?
Before an assistant reads your data, the data has to be right. We review your tracking and tell you what to fix first.
See analytics consultingWhich questions are worth asking?
The best questions are specific, tied to a decision and checkable. A vague request such as "how is my SEO?" returns a vague answer. Here are seven that work, with the reason each one is useful.
- "Which queries have many impressions but a click-through rate under 1% on page one?" These are pages that show up but are not chosen. The title or the description may not match what the searcher wanted. Rewrite the snippet, then measure again.
- "Which pages lost the most clicks between the last 28 days and the 28 days before?" A period comparison points you to the pages that need a look first. It does not say why, so open the page and the search results before deciding.
- "Which queries rank between position 4 and 15 for my product pages?" Pages just below the first results are often the cheapest to improve: a clearer answer, a stronger internal link, a missing section.
- "Which queries start with what, how, why or can, and which of my pages receive them?" This gives you the real questions people ask, in their own words, for your answer blocks. Our guide to answer engine optimization shows how to use them.
- "Do two of my pages receive impressions for the same query?" Two pages competing for one query is a sign of overlapping content. Decide which page owns the topic and link the other to it.
- "What is the index status of these 20 URLs?" Use the URL inspection tool on pages you just published or changed. Keep it to a short list because the inspection quota is lower than the performance quota.
- "How do mobile and desktop differ for my top ten pages?" A large gap in position or CTR between devices can point to a layout or speed issue worth testing.
A few habits improve every answer. Name the property, the date range and the metric. Ask for the numbers in a table, not only in a summary. Ask the assistant to state which filters it used, then compare one result with the Search Console interface to check that it matches.
How does this differ from the Search Console interface and a spreadsheet?
The interface is best for looking, the API is best for volume and the assistant is best for asking follow-up questions. Each has a row limit, and they are not the same.
Google's help documentation states that the tables in the Performance report can show a maximum of 1,000 rows, so some rows might be omitted. The API's Search Analytics method accepts a row limit from 1 to 25,000 per request (default 1,000) and a start row for paging. A server that uses the API can therefore return more rows than the interface, if it lets the assistant ask for them. Some servers cap the number on purpose, so check the README.
The assistant is not a replacement for a proper report. It is a faster way to ask the first twenty questions. When the same question comes back every week, that question belongs in a scheduled report, and our AI workflows service is where recurring tasks like this one are built.
Does Google offer an official Search Console MCP server?
We found no Search Console MCP server published by Google, as of October 2026. Searches of Google's developer documentation and of GitHub returned community projects only, and the Search Console API documentation we read does not mention MCP.
Absence of a result is not proof, so check again before you rely on this sentence. Google's pages and repositories change, and an official server could appear. For comparison, Google Analytics has a repository titled "Google Analytics MCP Server (Experimental)" under the googleanalytics organization on GitHub. Its README says that most tools are read-only, that one write tool needs an explicit flag and that it uses the analytics.readonly scope. Our guide to the Google Analytics MCP server looks at it in detail. That repository does not cover Search Console.
Which community servers exist?
Several community projects exist, written by independent developers, with different languages, tools and safety choices. Here are three we read on GitHub in October 2026. They are examples of the range, not endorsements, and we have not run them.
| Project | What the README says | Auth | Writes? |
|---|---|---|---|
ahonn/mcp-server-gsc | MIT license. One tool, search_analytics, with dimensions, filters and a quick-wins mode. Up to 25,000 rows per request. | Service account JSON key | Not stated |
AminForou/mcp-gsc | MIT license. Performance, URL inspection and sitemap tools. | OAuth (recommended) or service account | Yes, optional. Some destructive tools stay disabled unless a flag is set |
ricoarrigoni/gsc-mcp-server | MIT license. Described as "read-only by design". Nine tools including query_performance, top_pages, compare_periods and inspect_url. | OAuth or service account | No. It requests webmasters.readonly and refuses broader scopes |
Read the differences, not only the features. The first and the third are narrow, which is easier to review. The second does more, including sitemap changes, and that is exactly why it hides the riskiest actions behind a flag. If your goal is to read data, a read-only server is the sensible start.
How do you connect one safely?
Connect it as you would grant access to a contractor: a separate account, the smallest permission that works, a code review and a way to cut access. The MCP project's own security guidance supports each of these points.
- Choose a read-only server. Prefer one that requests
webmasters.readonly. Google's documentation lists it as a separate scope from the read/write one. - Read the code before you run it. The MCP security guidance warns that local servers can run arbitrary commands with the privileges of the client, and that a malicious startup command or payload is a real attack route. Check what the project installs, what it calls over the network and what it writes to disk. If you cannot read the code, ask someone who can.
- Check the exact start command. A configuration entry such as
npxplus a package name downloads and runs code. The MCP guidance recommends that a client show the exact command before it runs. Read it yourself even if your client does not. - Use a dedicated Google account or service account. Search Console lets an owner grant a person one of several permission levels. Google describes a full user as someone who can view all data and take some actions, and a restricted user as someone with simple view rights on most data. A restricted or full user on a dedicated identity limits the damage if a token leaks. A service account works the same way: add its email as a user on the property you want, and nothing else.
- Limit the properties. Grant access to one property first, not every site you manage.
- Keep credentials out of shared places. A key file or token should not sit in a synced folder, a repository or a chat. Store it with the permissions your operating system offers.
- Test with a question you can verify. Ask for last week's total clicks and compare with the interface. If the numbers differ, find out why before trusting anything else.
- Review and revoke. Remove the user from the property or delete the key when you stop using the server.
Scope minimization is a recurring theme in the MCP security guidance: tokens that carry broad scopes make a leak worse and revocation harder. The same logic applies here. If a server works with read-only access, there is no reason to hold more.
Want assistants connected to your data, with guardrails?
We design AI workflows with least-privilege access and a human review step. Tell us what you want to ask and we reply with a plan.
Explore AI workflowsWhat limits should you know about?
The assistant can only analyze what the API returns, and the API has retention, volume and quota limits. Knowing them stops you from drawing a wrong conclusion from a clean-looking answer.
- 16 months of history. A Google Analytics help page states that Search Console keeps data for the last 16 months. If you need longer trends, export and store the data on a schedule.
- Top rows, not every row. Google's documentation for the Search Analytics method says the API does not guarantee to return all data rows but rather top ones. A separate guide says the API exposes a maximum of 50,000 rows of data per day per search type, and that some data may be dropped when you request page and query together.
- Anonymized queries. Google's help page explains that not all queries are shown: it might not track queries made a very small number of times or those that contain personal or sensitive information. Totals by page can therefore be higher than the sum of the visible queries.
- Freshness. The newest data can be preliminary, meaning it is still being collected and might change in the next few hours. The API's default
finalstate returns only finalized data, so the last day or two may be missing unless the server asks forall. - Quotas. Google's usage limits page lists, for the Search Analytics query, 1,200 queries per minute per site and per user, and, for URL inspection, 2,000 queries per day and 600 per minute per site. It also notes that wider date ranges and grouping by page or query increase the load. Ask the assistant for focused requests, not for everything at once.
- Search Console is not Analytics. Google's help page notes that Search Console data may differ from Google Analytics, and that Search Console often covers only a subset of your site URLs. Treat each as a different measurement.
There is a final limit that is not in the API. The assistant writes fluent text, and a fluent summary can hide a wrong filter or a misread column. Ask it to show the request it made, and treat every number you will act on as something to verify.
Where do the assistant's answers go wrong?
Most errors come from the request, not from the data. The assistant may choose the wrong date range, group by the wrong dimension or mix up a page-level and a query-level total. It may also describe a correlation as a cause, such as saying a title change "caused" a rise when two other things changed the same week.
Three habits reduce the risk. First, ask the assistant to list its assumptions: property, dates, search type, filters. Second, keep a short list of control questions, such as "total clicks last 28 days", whose answers you know from the interface. Run them after any change to the setup. Third, keep a human in the loop for decisions. The assistant proposes a list of pages to improve. A person reads each page and the results, and decides.
For small sites, the rest of the work is the usual fundamentals. Our guide to SEO for small business and our overview of AI SEO cover what to do with the findings, and the SEO, GEO and AEO service page shows how these pieces fit together.
Does it work with Claude and other assistants?
It works with any assistant that supports MCP servers, and the setup steps differ by client. The MCP documentation lists Claude, ChatGPT, Visual Studio Code and Cursor among the supported clients. Each client has its own place to register a server, usually a configuration file or a settings screen. Our article on Claude and MCP covers the Claude side.
Frequently Asked Questions
What is a Google Search Console MCP server?
It is a program that connects an AI assistant to the Search Console API through the Model Context Protocol. The assistant can then call tools to fetch queries, pages, clicks, impressions, CTR, position, sitemaps and URL index status, and explain the results in plain language.
Is there an official Google Search Console MCP?
As of October 2026 we found none. Our searches of Google's developer documentation and of GitHub returned only community projects. Check Google's documentation again before relying on this answer, because the situation can change.
Is it safe to connect Search Console to an AI assistant?
It can be, if you limit the risk. Use a read-only server, read its code, use a dedicated account with access to one property, store credentials carefully and revoke access when you stop. A server you did not review runs code on your machine with your privileges.
Can the assistant change anything in my Search Console?
Only if the server and the permission allow it. The read-only scope does not permit changes. Some servers add optional tools for sitemaps or properties, usually behind a flag. If you only want answers, choose a server that requests the read-only scope.
How much data can it see?
Search Console keeps 16 months of data. For each request, the API accepts up to 25,000 rows, and it returns top rows rather than guaranteeing every row. Anonymized queries are not shown. The assistant can only work with what the API returns.
Do I need to be a developer?
Not to ask questions, but you need some comfort with installing a server, editing a configuration file and creating credentials in Google Cloud. If that is not your daily work, ask someone technical to do the setup and the code review.
What is the difference between this and the Google Analytics MCP server?
They cover different data. Search Console shows how people find your site in Google Search: queries, impressions, clicks and position. Analytics shows what visitors do on the site. Using both lets an assistant connect the two, but the numbers will not match, because each tool measures something different.
Next steps
Start with a decision, not a tool. Write down three questions you would like answered each month, then check whether the Search Console interface already answers them in a few clicks. If it does not, or if you want to ask follow-ups, try a read-only server with a test setup and compare results with the interface.
If you want to see where the connection fits in a wider measurement plan, our analytics work starts from clean data, and the pieces of an assistant-based workflow are part of our AI workflows offer. Our page on SEO, GEO and AEO explains what to do with the findings. When you are ready, get in touch and tell us what you want to ask.
Sources
Checked in October 2026.
- Model Context Protocol, "What is the Model Context Protocol (MCP)?": https://modelcontextprotocol.io/docs/getting-started/intro
- Model Context Protocol, "Security Best Practices" (local MCP server compromise, scope minimization): https://modelcontextprotocol.io/docs/tutorials/security/security_best_practices
- Google Search Central, Search Console API, Search Analytics query method (scopes, row limit, start row, data state, top rows): https://developers.google.com/webmaster-tools/v1/searchanalytics/query
- Google Search Console API, "Getting all your data" (50,000 rows per day per search type, dropped data): https://developers.google.com/webmaster-tools/v1/how-tos/all-your-data
- Google Search Console API, "Authorizing requests" (OAuth scopes): https://developers.google.com/webmaster-tools/v1/how-tos/authorizing
- Google Search Console API, "Usage limits": https://developers.google.com/webmaster-tools/limits
- Search Console Help, "Performance report" and data notes (1,000 rows in tables, anonymized queries, differences from Analytics): https://support.google.com/webmasters/answer/7576553 and https://support.google.com/webmasters/answer/96568
- Search Console Help, "Managing users and permissions": https://support.google.com/webmasters/answer/7687615
- Google Analytics Help, "Search Console data retention" (16 months): https://support.google.com/analytics/answer/13682862
- Google Analytics MCP Server (Experimental), GitHub: https://github.com/googleanalytics/google-analytics-mcp
- Community server
ahonn/mcp-server-gsc, README: https://github.com/ahonn/mcp-server-gsc - Community server
AminForou/mcp-gsc, README: https://github.com/AminForou/mcp-gsc - Community server
ricoarrigoni/gsc-mcp-server, README: https://github.com/ricoarrigoni/gsc-mcp-server
Newsletter
One short email when we publish
Notes on AI, image, video and brand work, from the people who do it. No spam.