SEO Cartograph · MCP

Search Console MCP: plug the GSC into your AI

Q: What is a Search Console MCP server?

An MCP (Model Context Protocol) server that exposes your Google Search Console data to an AI such as Claude. Instead of exporting a performance report and pasting it into a chat, the AI queries the property directly: real queries, pages, positions, impressions and clicks, by country and device. SEO Cartograph exposes Search Console, GA4, the crawl and rank tracking through the same connector.

Q: How far back can the AI look?

Google keeps roughly 16 months of Search Console data, and that is the ceiling. The Startup plan gives six months of it, Croissance the full history. Data is fresh to two or three days ago, which is Google's own consolidation delay.

Your AI queries your Google Search Console property directly: real queries, pages, positions, impressions and clicks, by country and device, across sixteen months of history. No export, no copy-paste, and on the same data as your GA4, through one connector.

Get started Pricing

Claude · SEO Cartograph MCP

Which queries started slipping this quarter, and on which pages?

Tools called

explore_search_console get_content_decay

dimension query · status down · 90 daysread only

The AI reads the queries Google actually showed you for

The protocol

What is a Search Console MCP server?

MCP (Model Context Protocol) is the open standard that lets an AI read data and call tools inside a third-party application. SEO Cartograph exposes Search Console through that channel: your AI phrases a question in plain language, the tool turns it into a Search Console API call, and the rows come back. One connector covers the GSC, GA4, the crawl and rank tracking.

The data

What your AI reads in Search Console

Exploration is open-ended: you pick the dimension, the date range and the filter, and the AI composes the query.

Dimensions

query
The real queries you surface on, the ones Google shows you for.
page
URLs, with their volume and average position.
country, device
Country and device, to isolate a market or mobile.
date
Day by day, to pin down when something turned.
searchAppearance
How the result appeared in the SERP.

Metrics

Clicks, impressions, CTR and average position on every row.
A live-or-dormant signal: days active out of 30, trend as a percentage, and a status among up, down, stable, new or dormant.
A summary mode that returns the aggregates without unrolling every row, for large properties.

Depth

Roughly sixteen months of history, Google’s own retention ceiling.
Fresh to two or three days ago, Google’s consolidation delay.
A filter on any dimension, a single country or a single page for instance.

Your real queries, as recorded by Google

This is the substantive difference with the keyword databases on the market. A third-party tool tells you what it believes you rank for, from a sample of the SERP. Search Console tells you what Google actually showed you for, how many times, and in which spot. On long-tail queries, local sites or a market like Switzerland, the gap between the two is often wide.

The tools

The Search Console tools in the connector

The badge shows the plan required.

`explore_search_console` Startup

Open-ended navigation inside the property. Any dimension, any range, with or without a filter. This is the tool for open questions: the queries for one page, the pages for one country, the day-by-day movement of a keyword.

`get_search_queries` Startup

Real queries over the last 28 days with impressions, clicks and position, filterable down to a single URL. The reflex when you are working on one page and want to know what it already surfaces on.

`get_cannibalisation` Startup

Queries where several of your pages rank, pair by pair, with a stability reading: Google alternates between your pages, or it has settled on one. Enough to decide which one to keep.

`get_content_decay` Startup

Pages losing ground, with the reason: clicks dropping, impressions dropping, position slipping, or the page gone from results altogether. The sort order that tells you what to rewrite first.

`get_analytics_history` Startup

How the site moves over time, weekly and monthly, with daily figures over 90 days. Search Console clicks and impressions arrive lined up with GA4 sessions and conversions, and with the top 3, 10 and 20 breakdown.

`inspect_url` Free

The real indexing verdict for a URL, with the canonical Google picked, the last crawl date and the mobile state. This is what lets the AI say “this page is not indexed” instead of guessing.

URL inspection queries Google and counts against the quota of 2,000 inspections per property per day set by Google.

Why it matters

Search Console and GA4 in the same question

A GSC-only connector answers “what position”. Combined with GA4 and the crawl, it answers “so what”.

A page loses clicks: the AI checks whether impressions dropped, whether the position slipped, or whether only the CTR got worse. Those three call for different fixes.
A query gains impressions without clicks: the title and meta description are the culprits, and the on-page editor takes over.
Two pages fight over the same query: cannibalisation is quantified, and internal linking settles it.
A page ranks well but does not convert: GA4 says so in the same conversation.

Setup

Plugging Search Console into your AI, in 3 steps

The Google connector is set up once in the dashboard, then the AI hooks into it.

Connect Search Console to the project

In the dashboard, using the Google account with access to the property.

Install the SEO Cartograph connector

One command through the Claude marketplace, or the MCP server URL for any other compatible client.

Ask your question

In plain language. The AI picks the dimension and the range, and hands you the rows.

Installation

MCP server URL

https://app.tnedjar.com/mcp/cartograph

Install through the Claude marketplace

claude plugin marketplace add tom-ned/seo-cartograph-mcp

claude plugin install seo-cartograph

In Claude Desktop: Settings, Plugins, Add a marketplace

Day to day

Questions worth asking once you are connected

Copy them straight into your MCP client.

Prompts

Which queries bring me impressions without clicks, and on which pages?

Show me the pages that lost ground over 90 days, and tell me why for each one.

On which queries are two of my pages competing with each other?

Which of my queries sit in positions 4 to 10, where one place changes everything?

Compare my clicks in the UK and the US over the last six months, by page.

Is this URL indexed, and which canonical did Google pick?

The same questions work in Claude.ai, Claude Desktop and Claude Code

Limits

What the Search Console connector does not do

Worth saying before you install it.

Google filters its own data

Rare queries are anonymised by Google, and the rows never add up to the property total. That gap comes from Google, and no tool closes it.

Sixteen months, no more

That is Google’s retention. Going further back requires snapshots taken along the way, which the dashboard history does once the project exists.

Fresh to two or three days ago

Google’s consolidation delay. For a check right after a page goes live, URL inspection answers immediately; performance data does not.

Your data stays yours

The point that matters when you plug an AI into a client’s data.

Performance data is read-only. The AI reads your reports, changes no property setting and removes no URL.
OAuth access with a consent screen. You see what the connector can read before you authorise it.
Data scoped to your account. The connector exposes only the projects in your workspace.
Revocable in one click. From your MCP client or from your workspace.

The intelligence is paid for by your Claude subscription

No AI surcharge on our side, no token meter, no credits to top up. You plug in the AI you already pay for, and SEO Cartograph supplies the data and the tools. The only meter that moves on our side covers live SERP analyses.

FAQ

Frequently asked questions

What is a Search Console MCP server?

An MCP server that exposes your Google Search Console data to an AI. Instead of exporting a performance report and pasting it into a chat, the AI queries the property directly: real queries, pages, positions, impressions and clicks, by country and device. SEO Cartograph exposes Search Console, GA4, the crawl and rank tracking through the same connector.

Which plan do I need to connect Search Console to my AI?

The MCP connector is available from the free plan, and Search Console data opens up on the Startup plan, at 15.-/month, with six months of history. Croissance, at 49.-/month, moves to full history, up to sixteen months. The pricing page compares all four plans. On the AI side you need a Claude plan that allows connectors.

Can the AI change my Search Console property?

Performance data is read-only. Two tools write, and only for what they announce: the URL inspection request, which queries the index, and the crawl request, which starts a SEO Cartograph analysis of your site. No property setting is changed and no URL is removed.

How far back can the AI look?

Google keeps roughly sixteen months of Search Console data, and that is the ceiling. The Startup plan gives six months of it, Croissance the full history. Data is fresh to two or three days ago, which is Google’s own consolidation delay.

Can the AI spot cannibalisation between my pages?

Yes. A dedicated tool returns the queries where several of your pages rank, with the query-and-page pairs, and says whether Google alternates between them or has settled on one. That is the signal you need to decide which page to keep and which to redirect or rewrite.

Are there extra AI credits to pay for?

No. The intelligence comes from the Claude subscription you already pay for. SEO Cartograph supplies the data and the tools, with no token meter and no AI credits to top up.

Which MCP clients are supported?

The connector follows the MCP standard. It is built and tested for Claude.ai, Claude Desktop and Claude Code, and any protocol-compliant client can connect through the server URL. The MCP connector page covers the setup and the full tool list.

Why plug Search Console into an AI rather than read the interface

The Search Console interface shows a thousand rows per report and leaves everything else to manual filtering. Crossing a query with the page that carries it, then with its movement over six months, takes three exports and a spreadsheet. With an MCP connector the question is one sentence and the tool composes the query. The difference shows up most on the slow work: finding the queries in positions 4 to 10 where one place changes the traffic, catching pages that slip before they disappear, and telling a position drop apart from a demand drop.

Cannibalisation, quantified instead of guessed

Two pages aiming at the same intent hold each other back, and the diagnosis is usually made on instinct. Here it rests on the query-and-page pairs Google recorded, with a reading of what Google actually does: alternate between your pages from one day to the next, or settle on one of them. Alternation says Google is hesitating and the call is yours. A settled choice says the second page earns nothing on that query. Internal linking then consolidates the one you keep.

One connector, several sources

Search Console is one source among several. The same MCP server reaches Google Analytics 4, the technical crawl, semantic clusters and rank tracking. That is what lets an SEO question get a complete answer, without switching tools or gluing exports back together. The features page covers the full scope, the SEO MCP page lists the tools, and the GSC and GA4 dashboard is the click-through version.