How to connect Claude to Google Analytics, and keep it read-only

Short answer: run Google’s open-source Google Analytics MCP server on your computer and add it to Claude Desktop or Claude Code, or add a hosted connector by its address in Claude’s Connectors settings. Either way, sign in with a Google user that can see your GA4 properties. Give it the analytics.readonly scope, and Claude can run reports but cannot change your property.

Which Claude are you using?

The route depends on where your Claude runs. A connector added in claude.ai is called from Anthropic’s cloud, so it can only reach a server at a public web address. A server that runs on your own machine needs Claude Desktop or Claude Code.

Plans and menu names from Anthropic’s help article on custom connectors, checked on 31 October 2026. The labels have moved before, so look for Connectors if yours read differently.

Route 1: Google’s open-source server on your computer

Google publishes the code at github.com/googleanalytics/google-analytics-mcp. It is read-only unless you switch writing on, and it is the cautious option. The tools it lists are further down.

1. A Google Cloud project with two APIs enabled: the Google Analytics Admin API and the Google Analytics Data API. GA4 has no developer token and no access level to apply for.
2. Credentials with the read-only scope. The README uses Application Default Credentials: gcloud auth application-default login with your OAuth client file and two scopes, https://www.googleapis.com/auth/analytics.readonly and https://www.googleapis.com/auth/cloud-platform. The command prints where it saved the credentials file. Keep that path.
3. Python 3.10 or newer, and pipx, the tool that runs the server without installing it for good.
4. The server in Claude’s config. In Claude Desktop, open Settings > Developer > Edit Config and add the block below, then restart the app.

claude_desktop_config.json, with your own path and project

{
  "mcpServers": {
    "analytics-mcp": {
      "command": "pipx",
      "args": ["run", "analytics-mcp"],
      "env": {
        "GOOGLE_APPLICATION_CREDENTIALS": "/path/to/application_default_credentials.json",
        "GOOGLE_CLOUD_PROJECT": "your-cloud-project-id"
      }
    }
  }
}

The README shows this block for Gemini; Claude Desktop reads the same mcpServers format.

In Claude Code

The same server is one command: claude mcp add analytics-mcp --scope user -e "GOOGLE_APPLICATION_CREDENTIALS=/path/to/file.json" -e "GOOGLE_CLOUD_PROJECT=your-project" -- pipx run analytics-mcp.

Check that it works

To check it works, ask Claude which Google Analytics properties it can see. That calls get_account_summaries, and an empty list means the Google user behind the credentials has no access to any property.

And from claude.ai or your phone?

Then the server has to live at a public address. Google also runs one for the Data API at https://analyticsdata.googleapis.com/mcp/v1. On 31 October 2026 it answered a connection and listed four tools, all marked read-only: run_report, get_metadata, run_realtime_report and check_compatibility. Its reference page in Google’s developer docs returned an error the same day, so I have not written up how to sign in to it from claude.ai. Until Google documents it, treat it as one to watch, not one to set up.

What Google’s server can do

Nine tools, and only one of them writes:

So out of the box, Claude can list your properties, read their setup and run reports. To let it add annotations, you start the server with --allow-write and sign in with the edit scope as well.

Route 2: a hosted connector

A vendor runs the server, and you paste its address into Claude. On Pro or Max: Customize > Connectors > + Add > Add custom connector, a name, the address, then Add. Claude finds the sign-in on its own, and you sign in with the Google account that sees your GA4 properties. In each conversation, switch the connector on from the + menu.

Before you click Allow on Google’s screen, check three things:

• The scope it asks for. “See your Google Analytics data” is the read-only one. If the screen also asks to edit your Google Analytics management entities, the connector can change your property.
• Whose app you are signing into. The consent screen names the app. It should match the vendor you chose.
• Where your data goes. A hosted server sees every answer it sends to Claude. If you manage clients, that is their data, not yours.

Climent Ads has its own server in the hosted app. It is in testing, behind OAuth, and it reads the data already imported into the console rather than calling Google live.

Route 3: build your own

If you want your own tools, you need the same Cloud project, the two APIs, an OAuth client and a consent screen. A public app needs Google’s verification first. MCP vs API covers what changes once the model picks the calls.

What Claude can change once it is connected

GA4 differs from Google Ads here. The Google Ads API has one scope, and it writes. Google Analytics has a read-only scope, so you can lock it at the sign-in. There are three places to keep Claude reading:

For a client’s property, the cleanest setup is a Google user with the Viewer role on that property, and you sign in to the connector with that one.

Reading has its own traps. Claude writes the report request itself, so check the date range, the dimensions and the filters before you trust a total. The last two or three days are still filling in. And GA4 counts sessions and conversions its own way, so its numbers will not match the ad platforms’. Reading Google and Meta together covers what to line up first.

Questions people actually ask

Related reading

• How to connect Claude to Google Ads. Google’s server, a hosted connector or your own, and why the access level does the locking there.
• How to connect Claude to Meta Ads. Meta hosts its own connector, so the setup is shorter, and the permissions matter more.
• Google Analytics MCP: the options. Official, third-party and ours, and what each one reads.
• Marketing attribution. Why GA4 and the platforms never agree on conversions.
• Connect your AI to your ad data. The same choice across Meta, Google and GA4, for any assistant.
• What is MCP. The standard behind every connector, in plain words.

Written by Manuel Climent · Last updated: October 2026