Google Analytics MCP Server 1.0: What It Does and How to Set It Up

What Google Released

Google’s official Google Analytics MCP server reached version 1.0.0 on 3 November 2026, and version 1.1.0 followed on 5 November. It is an open-source connector, published by the Google Analytics team on GitHub. It lets an AI assistant query your GA4 properties directly.

MCP stands for Model Context Protocol. It is an open standard that gives AI assistants a safe, structured way to use outside tools. You don’t export a CSV and paste it into a chat. You ask a question, the assistant calls the GA4 Data API or Admin API for you, and then it explains the result.

• Repository: googleanalytics/google-analytics-mcp on GitHub.
• Package: analytics-mcp on PyPI. It needs Python 3.10 or newer.
• Status: version 1.0.0 “graduated” the project out of its experimental label. That came after more than a year of public work since mid-2025.

What Changed in 1.0 and 1.1

The two November releases turned a handy reporting add-on into something teams can really roll out. Here’s what each one brought.

Some early social posts call the server “experimental” and “read-only except for annotations”. Both were true a few weeks ago. As of 1.1.0, both are out of date.

What It Can Do Today

The server offers a set of named tools. Your assistant picks which one to call based on your question. So you never have to remember the tool names yourself.

Reporting

• Standard reports (run_report): any mix of dimensions and metrics, with filters, sorting and date comparisons. Each response can return up to 250,000 rows, with paging for bigger pulls.
• Realtime reports (run_realtime_report): what’s happening on the site right now. Since 1.1.0 you can ask for set minute windows, such as “the last 5 minutes” versus “the 25 minutes before that”.
• Funnel reports (run_funnel_report): where people drop off, step by step. You can break it down by device, country or browser, or by user segment. A “next action” view shows which events or pages people go to after they leave or finish the funnel.
• Conversions report (run_conversions_report): ad costs, conversions, return on ad spend and attribution models in one call.

Auditing and Property Checks

• Account and property details: list every account and property you can access, and read a property’s settings.
• Key events (list_key_events): see which events count as key events, including how they’re counted and any default value.
• Custom definitions: read all custom dimensions and metrics on a property.
• Google Ads links: check which Ads accounts are linked. In my experience, a missing link is a common cause of “missing” conversion data.
• Annotations: read the notes on your reporting timeline.

Setup Changes (Only When You Turn on Write Access)

• Create annotations to mark launches, bid changes or tracking fixes on the timeline.
• Create custom dimensions at event, user or item level (needs the Editor role).
• Mark an event as a key event (needs the Marketer or Editor role).

There are no tools to delete or archive anything. I think that’s a sensible limit for a first version. It means the worst a bad prompt can do is add something you then remove by hand.

Read-Only by Default: How Write Access Works

Out of the box, the server can only read data. Write tools only appear when two things are both true. The server must be started with the --allow-write flag. And your Google credentials must include the analytics.edit scope on top of the usual analytics.readonly scope.

Your role on the property still applies. If you’re a Viewer in GA4, the assistant can’t create a key event for you, whatever the flag says.

• For analysts and agencies: keep a read-only setup for everyday questions. Only turn on write access on a separate setup, on purpose. This is how I’d run it.
• For client properties: agree in writing which changes the assistant may make. Annotations are low risk. Custom dimensions are limited per property and can’t be renamed freely, so treat them with care.

One small mismatch to know about. Google’s developer documentation page still says the server only handles read requests. But the repository’s README, which comes with the code, documents the opt-in write tools. Go by the README for what the code can really do.

What You Need to Set It Up

This isn’t a one-click plugin yet. Expect 20–40 minutes the first time if you haven’t used Google Cloud before.

1. A Google Cloud project with the Google Analytics Admin API and Google Analytics Data API turned on.
2. Python 3.10+ and pipx. pipx runs the server without a manual install.
3. Credentials: Application Default Credentials made with the gcloud command line. You can sign in as yourself through an OAuth client, or act as a service account (a Google account for software, not a person). The account needs access to the GA4 properties you want to query.
4. An assistant that supports MCP: the README gives ready-made instructions for Gemini CLI, Gemini Code Assist and Claude Code. Other MCP clients can use the same command and environment variables.
5. Add the server to your assistant’s settings, pointing it at your credentials file and project ID. Only add --allow-write to the arguments if you want the write tools.

On macOS, a few people hit DNS errors when calling the APIs over a VPN. The project’s troubleshooting guide fixes this by setting GRPC_DNS_RESOLVER=native.

Local or Hosted: Choosing How to Run It

You can run the server on your own computer, or host it once for a whole team.

The hosted mode matters most for agencies. Each person only gets answers from properties their own account can see. So you don’t share one powerful service account across the team. If I were setting this up for a team, this is the mode I’d pick.

Prompts That Work Well

Be clear about the property, the date range and the comparison you want. These are good starting points:

Paid Media Checks

• “For property 123456, compare sessions, key events and revenue by Google Ads campaign for the last 30 days against the previous 30 days.”
• “Show the conversions report with cost and ROAS by campaign for October, using data-driven attribution.”
• “Is traffic from the new Performance Max campaign arriving right now? Compare the last 5 minutes with the 25 minutes before.”

Funnels and User Journeys

• “Build a checkout funnel (view_item, add_to_cart, begin_checkout, purchase) for paid traffic only, broken down by device. Where’s the biggest drop?”
• “For people who abandoned at begin_checkout, what did they do next?”

Audits

• “List every key event on this property with its counting method, and flag any that look like duplicates.”
• “Which custom dimensions exist, and which of them had no data in the last 90 days?”
• “Which Google Ads accounts are linked to this property?”

Housekeeping (Write Access On)

• “Add an annotation for today: switched the Search campaigns to a target ROAS of 400%.”
• “Create an event-scoped custom dimension for the plan_tier parameter.”

Limits and Things to Check

The assistant is only as good as the question and the data behind it. A few habits keep the answers trustworthy:

• Check the first few answers against the same report in the GA4 interface. Differences usually come from a different date range, attribution setting or filter, not from a bug. I never trust a new data tool until it matches GA4 on a few known numbers.
• Watch API limits. Very large or repeated requests use the same Data API quota as any other connection, including dashboards.
• Thresholding and sampling still apply. If GA4 hides small numbers for privacy, the assistant sees the same gaps.
• Funnels use an alpha API. The funnel tool relies on Google’s alpha (early test) funnel reporting endpoint, so expect changes now and then.
• Data goes to your AI provider. Report results are sent to whichever assistant you use. Check that this fits your privacy terms and any client contracts.

What It Means for SEO and Paid Search Teams

The biggest win is speed on routine questions. Weekly checks that used to need a dashboard build or an export now take one sentence. That leaves more time for actually fixing things.

• SEO: quickly split organic landing-page results by device or country after an update. Add an annotation on the date a core update or site change went live.
• Paid search: check campaign launches in realtime, and compare ROAS by attribution model without leaving your chat window.
• Analytics hygiene: audit key events and custom dimensions across many properties far faster than clicking through each one.
• AI traffic: pair it with our guide on getting cited in AI Overviews and ChatGPT search to track how AI referrals grow over time.

If you’re also cleaning up a site after a ranking drop, our technical SEO audit checklist and core updates tracker show which date ranges are worth comparing.

Next Steps

1. Create or pick a Google Cloud project and turn on the Analytics Admin and Data APIs.
2. Install pipx and set up Application Default Credentials with the read-only scope.
3. Add the server to Gemini CLI or Claude Code using the README’s example settings.
4. Ask three questions you already know the answers to, and compare them with GA4.
5. Only then decide if your team needs write access or the hosted mode for many users.

My advice: start small and read-only. The Google Analytics MCP server is most useful once you trust its answers, and that trust comes from checking it against GA4 first.