How to Connect Claude to Google Analytics and Search Console With MCP (Step by Step)

TL;DR: You can ask Claude plain English questions about your Google Analytics 4 and Search Console data by adding one remote MCP server as a custom connector. Setup takes four clicks in Claude, one command in Claude Code, and the access is read only. Below are the exact steps, the permissions you grant, example questions, and how to handle ads data.

If you manage SEO or ads, you probably spend a fair part of the week moving between Analytics, Search Console and a spreadsheet to answer questions your boss or client could ask in one sentence. Which pages lost clicks? Where did the traffic come from? What changed after the last update?

MCP lets an AI assistant read that data directly, so you can ask those questions in a chat. This guide shows how to set it up with a free, read only server, and what to check before you trust the answers.

What MCP is, in one paragraph

MCP, the Model Context Protocol, is an open standard for connecting AI applications to external systems. Think of it as a standard plug between an assistant, such as Claude or ChatGPT, and a data source. Instead of exporting a report and pasting it into a chat, the assistant asks the server for the numbers when you ask a question.

A remote MCP server lives at a web address. You give that address to your AI app, sign in once, and approve what it may read.

What you need before you start

• A Google account that has access to your Google Analytics 4 property and your Search Console site.

• An AI app that supports custom connectors or remote MCP servers. Claude on the web and desktop, Claude Code and Cursor all do. ChatGPT supports it through developer mode, and the available options depend on your plan.

• The server address. We use a free analytics MCP server whose address is https://audit.greatinflux.com/mcp.

Claude's own help center has a general guide to custom connectors with remote MCP, which is worth a look if your plan or workspace has admin controls.

Connect it in Claude (web or desktop)

1. Open Settings and go to Connectors.

2. Choose Add custom connector, name it (we call ours "ONE Dashboard"), and paste the server address.

3. Click Connect, sign in with your Google account when prompted, and click Allow.

4. In a new chat, open the + menu, switch the connector on, and ask your question.

That is the whole setup. If the connector does not appear in the + menu, check that it is switched on for the chat you are in.

Connect it in Claude Code

If you work in a terminal, add the server with one command:

claude mcp add --transport http one-dashboard https://audit.greatinflux.com/mcp

Then type /mcp, pick the server, and choose Authenticate. A browser window opens for the Google sign in.

Connect it in Cursor or ChatGPT

In Cursor, go to Settings, then MCP, choose Add new MCP server, and add an entry named one-dashboard with the same address.

ChatGPT needs developer mode turned on, then a connector that uses OAuth for sign in. Because custom connectors depend on your plan, check what your account allows before you spend time on it.

What access you are actually granting

This part deserves attention, because connecting an assistant to business data is a real decision.

The server requests read only Google permissions: analytics.readonly and webmasters.readonly, plus basic profile information so it can identify you. It cannot change your Analytics or Search Console settings or data. It reads the same reports you would see in a dashboard, and only when you ask a question.

You can remove access in three ways: delete the connector in your AI app, disconnect from the connections page of the dashboard, or revoke it in your Google Account permissions. Run through that list once, so you know the exit before you need it.

A general rule worth following with any MCP server: read the scopes on the Google consent screen, and say no if they ask for more than the job needs. Write access to a live ad account is a different level of risk from read access to reports.

Questions worth asking first

Once connected, start with questions you can already check yourself. That is how you learn whether to trust the answer.

• "Which pages lost the most clicks in Search Console over the last 28 days compared with the previous 28?"

• "What are my top ten landing pages by sessions this month, and how has each changed since last month?"

• "Which queries get lots of impressions but few clicks, and what page ranks for each?"

• "Where did my organic traffic grow or fall by country?"

• "Summarize what changed in my traffic this week in five bullet points for a client."

Pick one you already know the answer to, and compare. If the assistant's numbers match your dashboard, you can lean on it for questions you have not yet asked.

What about Google Ads and Meta Ads?

A fair question, since many marketers want ad data in the same chat. The server described here connects to Analytics and Search Console. It does not document a direct Google Ads or Meta Ads connection.

You can still see ad performance through Analytics, as long as the ads are tagged properly. Google Ads can feed GA4 through its native link, and Meta Ads traffic shows up when your ads carry UTM parameters. Then you can ask about sessions, conversions and landing pages by campaign.

That gives you the traffic and outcome side of the story. Spend, bids and ad level detail still live in the ad platforms, so keep those reports open for budget decisions.

Tips for getting reliable answers

• Be specific about dates. "Last 28 days versus the previous 28" beats "recently."

• Ask for the table first. Request the numbers, then ask for interpretation. It makes errors easier to spot.

• Check one number. Open the real report and confirm at least one figure per session.

• Watch for sampling and delays. Analytics data can lag, and recent days may change.

• Keep a client safe. If you are connecting a client's account, get their approval and use their own Google sign in.

One connector, fewer tabs

The real gain is not a clever trick. It is that questions that used to take four tabs and a spreadsheet now take a sentence. You still need judgment about what the numbers mean, and you still check the source. But the first draft of the answer arrives in seconds.

Set it up once, test it on a question you know the answer to, and add the harder questions after that. Teams that report to clients can also save a standard set of prompts and rerun them each month, which keeps the reporting consistent.

Connect the server to one Google account this week, and compare its answer to a report you already trust.