GUIDE · FOR EVERYONE NEW TO MCP

What Is a Search Console MCP Server? A Plain-English Guide

A Search Console MCP server lets an AI assistant read your real Google and Bing search data on request. Here is what that means, what it can answer, where its limits are, and who it is genuinely useful for.

By GSC MCP Server teamUpdated 9 min read

The short answer

A Search Console MCP server is a small service that sits between your Search Console account and an AI assistant such as Claude, ChatGPT or Cursor. When you ask a question like “which pages lost clicks this month?”, the assistant calls the server, the server queries the Search Console API with your permission, and the assistant answers using the numbers that came back.

Nothing about your data changes. Search Console is still the source of truth. What changes is the interface: instead of filtering the Performance report, exporting 1,000 rows to a spreadsheet and pasting it into a chat, you ask in plain English and get an answer grounded in live data.

If you just want one connected, GSC MCP Server is our hosted Google Search Console MCP server; this page explains how the category works.

What MCP actually is

The Model Context Protocol (MCP) is an open standard, introduced by Anthropic in late 2024, for connecting AI applications to outside tools and data. Think of it as a universal adapter. A service describes the tools it offers (their names, what they do, and which inputs they accept), and any MCP-compatible assistant can discover and call them.

Before MCP, every AI app needed its own custom integration for every data source. With MCP, one server works across Claude, ChatGPT, Cursor, VS Code, Devin Desktop, Cline and others. That is why a single Search Console MCP server can serve a marketer in ChatGPT and a developer in Cursor at the same time.

How a request flows

  1. You ask your assistant a question about your site.
  2. The assistant decides it needs data and picks a tool, for example get_search_queries, with inputs such as a date range and property.
  3. The MCP server checks your authorization and calls the Search Console or Bing Webmaster API.
  4. The server returns structured results: clicks, impressions, CTR, average position and so on.
  5. The assistant reads the results and answers, often calling more tools to dig deeper.

MCP vs the API vs CSV exports

All three read the same Search Console data. They differ in who does the work:

  • CSV export. The Performance report exports at most 1,000 rows per table, for one property and one set of filters at a time. It is a static snapshot you then analyse yourself.
  • The Search Console API. Returns up to 25,000 rows per request, with paging, across the full 16 months. You write the code, create the OAuth client and handle the quotas.
  • An MCP server. Wraps those API methods in named tools an assistant can call. You ask a question; the assistant picks the calls, chains them and explains the result. It has the same data and the same limits as the API, without the code.

The problem it solves

Search Console is the most accurate source of organic search data you have, and it is free. It is also slow to interrogate. A few everyday frustrations:

  • The 1,000-row ceiling. The Performance report and its exports show at most 1,000 rows. Mid-sized sites routinely have tens of thousands of queries, so the long tail stays hidden unless you use the API.
  • Questions that span filters. “Which queries moved from page two to page one, on mobile, in the UK?” takes several filter changes, a comparison and some spreadsheet work.
  • Multiple properties. Consultants and agencies switch between dozens of properties, and Google and Bing data live in two different tools.
  • Interpretation. Even after you have the numbers, turning them into a prioritised plan is a separate job.

An MCP server handles the fetching through the API (which reaches far beyond 1,000 rows), and the assistant handles the slicing, comparing and explaining.

What you can ask

These are typical questions people run in their first week. Each one maps to one or more tool calls.

Try this prompt
Compare the last 28 days with the previous 28 days for my main site. Which ten pages lost the most clicks, and what queries drove the drop?
Try this prompt
Show queries where I rank between positions 8 and 20 with at least 500 impressions. Group them by landing page and suggest what to improve on each page.
Try this prompt
Is /pricing indexed? What canonical did Google choose, and when was it last crawled?
Try this prompt
For every client account, give me clicks, impressions and CTR for last month and the change from the month before, as a table.

For deeper versions of these workflows, see our guides on striking distance keywords and diagnosing a traffic drop, or copy more Search Console prompts.

The API methods behind the tools

Whatever a Search Console MCP server calls its tools, its read tools use four Search Console API methods. Knowing them tells you what any server can and cannot answer.

Typical MCP toolAPI methodWhat it returns
List propertiessites.listThe properties your Google account can access, with your permission level
Performance data (overview, top queries and pages, period comparisons, opportunities, cannibalisation)searchanalytics.queryClicks, impressions, CTR and average position, grouped by query, page, country, device, date or search appearance
URL inspectionurlInspection.index.inspectIndex verdict, coverage state, last crawl, canonicals, robots.txt state and rich results for the indexed version of one URL
Sitemapssitemaps.listSubmitted sitemaps with submission and download dates, errors, warnings and submitted URL counts (no indexed counts; inspect pages individually)

Servers that also cover Bing call the Bing Webmaster Tools API for the same jobs. In GSC MCP Server, list_search_properties, inspect_url and list_sitemaps map to sites.list, urlInspection.index.inspect and sitemaps.list. query_search_analytics and the shortcuts built on the same method (get_search_overview, compare_search_periods and others) cover performance data. get_bing_crawl_stats adds Bing crawl data, and multicall runs up to ten calls in one request.

MCP vs exports, Looker Studio and SEO suites

An MCP server does not replace every tool you use. It fills a specific gap: fast, conversational access to first-party search data.

ApproachStrengthWeakness
Search Console UI and CSV exportFree, official, no setup1,000-row cap, manual filtering, one property at a time
Looker Studio dashboardsGood for recurring visual reportsAnswers only the questions you designed it for; slow to change
Third-party SEO suitesCompetitor data, backlinks, rank trackingRankings and traffic are estimates, not your first-party clicks
BigQuery bulk exportKeeps every day from the day you enable it, beyond 16 months (no backfill)Requires SQL and a Google Cloud project
Search Console MCP serverAd-hoc questions in plain language, Google + Bing togetherBound by API limits and the 16-month retention window

Many teams pair them: an SEO suite for competitive research, a dashboard for stakeholders, and an MCP server for the daily “why did this happen and what should I do?” questions. We compare the last pairing in Search Console MCP vs Ahrefs and Semrush MCP.

Hosted vs self-hosted servers

A self-hosted server runs on your own machine or cloud. You create a Google Cloud project, enable the Search Console API, set up an OAuth client or service account, and point your client at the server. It costs nothing but setup time and upkeep. Claude and ChatGPT in the browser can only reach a server with a public HTTPS URL, so local servers suit desktop apps, IDEs and terminal agents.

A hosted server runs for you. You sign in with Google, paste one URL into your assistant, and the vendor stores the OAuth tokens and keeps the server running, usually for a monthly fee. We compare the open-source and hosted Search Console MCP servers side by side.

Who gets the most out of it

  • Freelance SEO consultants who manage several sites and want to move from data gathering to recommendations faster.
  • Agencies producing monthly reports across many client properties. See our agency reporting workflow.
  • SaaS and startup founders who own SEO alongside ten other jobs. Our 30-minute weekly routine is written for you.
  • Content marketers and editors who need to know which articles to refresh and which overlap. Start with keyword cannibalization.
  • Developers who want indexing and canonical checks next to their code. Read technical SEO in Cursor and VS Code.

The read-only scope: webmasters.readonly

Google offers two OAuth scopes for Search Console: webmasters, which can also submit and delete sitemaps and add or remove properties, and webmasters.readonly, which can only read. Check which one a server asks for on the Google consent screen. GSC MCP Server requests only webmasters.readonly.

You connect Google and Bing through their own consent screens, so you never hand over a password. AI clients then connect to the server with their own scoped, revocable authorization, so you can disconnect ChatGPT without disconnecting Claude.

Privacy: what your AI provider sees

Your AI provider will see the data the assistant retrieves during a conversation, the same as if you had pasted a report into the chat. If your organisation has rules about which assistants may handle client data, apply them here too. Teams with stricter requirements can self-host the open-source server once it is released; that release is planned, not yet available. Our privacy policy explains what the hosted service stores.

Limits worth knowing

An MCP server inherits the limits of the APIs underneath it. These are the ones that shape answers most often:

  • 16 months of history. Search Console keeps performance data for 16 months. Year-over-year comparisons work; two-year trends do not, unless you archive data yourself.
  • Anonymised queries. Google omits queries searched by very few people to protect privacy, so query-level totals are usually lower than page-level totals. This is expected, not a bug.
  • Data freshness. Data typically lags by two to three days, and very recent days may change (see Google’s About Search Console data).
  • URL Inspection quota. Google allows 2,000 inspections per property per day, so inspect important URLs rather than entire sites.
  • AI features in the totals. Impressions and clicks from AI Overviews and AI Mode are counted inside the Web search type of the main Performance report. Since 31 August 2026 every site has a separate Generative AI report (announced in June), with AI impressions by page, country, device and date. It is not available through the API, so an MCP server cannot read it; Google’s help page covers what it includes.

How to get started

  1. Create a free account with Google or GitHub.
  2. Connect a Google Search Console or Bing Webmaster Tools account and pick the properties you want available.
  3. Add the server to your assistant. Our setup guide for Claude, ChatGPT, Cursor and VS Code walks through each client.
  4. Ask a first question, such as “What changed in my search traffic this month?”

Current plans, including the free one, are on our pricing section.

Frequently asked questions

How is an MCP server different from the Search Console API?

An MCP server is a layer on top of the API. The API returns raw data to code you write; an MCP server wraps the same API methods in named tools that an AI assistant can discover and call, so you ask in plain language instead of writing requests. The data, the 16-month window and the quotas are the same.

Does an MCP server change my Search Console data?

Not if it uses the read-only scope. GSC MCP Server asks Google only for webmasters.readonly, so it can read performance data, sitemaps and URL Inspection results but cannot submit or delete sitemaps, add or remove properties, or change settings.

Which AI assistants support MCP?

Claude (web, desktop and Claude Code), ChatGPT through custom MCP plugins, Cursor, VS Code with GitHub Copilot, Codex, Antigravity CLI, Devin Desktop (formerly Windsurf), Cline and a growing list of other clients support MCP. Setup differs slightly between them.

Can the AI make up numbers?

Any language model can misstate figures, which is why tool calls matter: the assistant fetches the numbers from Search Console instead of guessing. Good practice is to ask it to show the table it used and to spot-check important figures in Search Console.

Do I need to be technical to use it?

No. With the hosted version you sign in, connect Google or Bing, and paste one URL into your AI app. An open-source, self-hosted release for teams that want to run the server themselves is planned.