Google Analytics MCP server: run Google's official one yourself, or have it hosted

Key points

• Google's official server lives on GitHub as googleanalytics/google-analytics-mcp. It is open source under the Apache 2.0 license, and its README marks it experimental.
• It runs locally and needs pipx, a Google Cloud project with the Analytics Admin API and Analytics Data API turned on, and Application Default Credentials with the analytics.readonly scope.
• ChatGPT developer mode adds a remote MCP server, and Claude reaches a custom connector's server from Anthropic's cloud, over the public internet.
• Mitos runs Google's official server itself, without changing its code. You sign in with Google, choose the GA4 properties you want and add one address to ChatGPT or Claude.
• Either way the server only reads. Google says it cannot change your Analytics settings, and Mitos connects with the read-only analytics.readonly scope by default.

Running Google's server yourself or having it hosted

Google's official Google Analytics MCP server

Google's developer documentation presents the Google Analytics MCP server as a way to connect your Analytics data to an LLM such as Gemini, to chat with your data or build your own agents. The code lives in the googleanalytics organization on GitHub under the Apache 2.0 license, and the README title marks it experimental.

The server uses the Google Analytics Admin API and the Google Analytics Data API. Google notes that it handles read requests only and cannot edit your Google Analytics configuration or settings. The README lists these tools:

• get_account_summaries: the user's Google Analytics accounts and properties.
• get_property_details and list_google_ads_links: details of one property and its links to Google Ads accounts.
• run_report and run_funnel_report: reports and funnel reports through the Data API.
• get_custom_dimensions_and_metrics: a property's custom dimensions and metrics.
• run_realtime_report: realtime reports.

What it takes to run it

The server runs on your own computer. The README gives these steps:

• Install pipx. The package needs Python 3.10 or later.
• Turn on the Google Analytics Admin API and Google Analytics Data API in a Google Cloud project.
• Set up Application Default Credentials for a user with access to your Google Analytics accounts or properties, with the analytics.readonly scope, for example through gcloud and an OAuth client.
• Register the server in your AI assistant's settings. The README shows Gemini CLI, Gemini Code Assist and Claude Code, which start it with pipx on the same computer.

Why a local server does not go into ChatGPT or Claude

For ChatGPT developer mode, OpenAI describes creating an app for your remote MCP server, over SSE or streaming HTTP. Anthropic writes that Claude connects to a custom connector's server from Anthropic's cloud rather than from your device, in every Claude app. That server must be reachable over the public internet.

So a server running on your computer is not an address you paste into ChatGPT or Claude on the web. Claude Desktop can run local MCP servers from its configuration file, but Anthropic describes that as a separate mechanism that claude.ai does not have.

How Mitos hosts it

Mitos runs Google's official Google Analytics MCP server on its own servers, without changing its code, and adds what a hosted service needs: Google sign-in, access to only the properties you choose, organizations and a way to revoke access.

It passes the server's read tools to ChatGPT or Claude through an allowlist, so a tool added in a later version of Google's server stays hidden until Mitos has reviewed it. You install nothing and create no Google Cloud project. Setup is self-serve at app.sar-soft.com and takes a few minutes. If you get stuck, email [email protected].

• Sign in with Google, allow read-only access (analytics.readonly) and choose the GA4 properties you want, each for one of your sites.
• Add https://mcp.sar-soft.com/mcp as a custom connector in ChatGPT (with developer mode turned on) or Claude.
• Sign in to Mitos when asked and authorize one organization. The connector reaches no other organization.
• Revoke the AI assistant in the Mitos app, and it takes effect within about a minute.

When each fits

Running Google's server yourself fits a developer who is comfortable with Google Cloud and credentials, and who works in a local AI assistant such as Gemini CLI or Claude Code.

Having it hosted fits a business owner or agency working in ChatGPT or Claude who wants to pick their properties without running anything. An agency creates one organization per client.

Either way, the AI assistant writes the answer, and it can be wrong or incomplete. Check the numbers that matter in GA4.

Questions to try

• How much traffic did we get in the last 30 days, compared with the 30 before?Sources: Google Analytics
• Which landing pages brought in the most visitors this month?Sources: Google Analytics
• Where did our visitors come from last week?Sources: Google Analytics
• Which pages get plenty of search impressions but little actual traffic?Sources: Search Console · Google Analytics

Questions people ask

References

• GoogleTry the Google Analytics MCP server
• Google Analytics (GitHub)google-analytics-mcp
• OpenAIDeveloper mode
• AnthropicGetting started with custom connectors using remote MCP