You want to ask Claude how your site is doing, without opening GA4 and building an exploration. There are two MCP paths today, and which one fits depends on whether you want to run Python and gcloud.
1. Google's official Google Analytics MCP server
Google maintains an open source MCP server for Analytics: https://github.com/googleanalytics/google-analytics-mcp
Per its README, it is written in Python and you run it with pipx, with a Google Cloud project that has the Analytics Admin API and Data API enabled, and credentials from gcloud auth application-default login. It runs reports, realtime reports and funnel reports, reads account and property details, and, if you pass --allow-write with the edit scope, it can also create annotations, custom dimensions and key events.
If you already work with gcloud, or you want the funnel report, this is the one.
2. A remote server: paste a URL, sign in with Google
The Google Analytics connector from Agents (by Ciro Cesar Maciel, who writes this) is for people who do not want to install anything. No Python, no gcloud, no Cloud project:
https://google-analytics.mcp.myagents.click/mcp
It gives your assistant seven tools, all read only:
| Tool | What it answers |
|---|---|
list_properties |
The GA4 accounts and properties your account can see |
run_report |
Any mix of dimensions and metrics, with period, filter and order |
compare_periods |
The current period against the previous one, with the change of each metric |
top_pages |
Most viewed pages, with users, engagement time and key events |
traffic_sources |
Channels and source/medium, with sessions, engagement and key events |
realtime |
How many people are on the site in the last 30 minutes, and where |
list_metrics |
The valid names on the property, custom ones included |
Ask in your own words:
- "Where did traffic come from this week, and what changed from last week?"
- "Which pages lead to the most key events?"
- "How many people are on the site right now, and from where?"
If you use custom metrics or dimensions, Claude finds the right names with list_metrics before building the report.
In Claude
- Settings, Connectors, Add custom connector.
- Name it "Google Analytics", paste the URL and click Connect.
- Sign in to the Agents account once, then choose the Google account with access to your GA4 properties.
- Start with "which properties can you see?" to check the right account came in.
In Claude Code and Cursor
claude mcp add --transport http google-analytics https://google-analytics.mcp.myagents.click/mcp
Cursor, in .cursor/mcp.json:
{
"mcpServers": {
"google-analytics": { "url": "https://google-analytics.mcp.myagents.click/mcp" }
}
}
Honest notes
- Google shows an "unverified app" notice when you sign in. Google is still verifying the Agents project for Analytics, which is a sensitive scope; until then you choose to continue, and the project has a cap of 100 users. If your team does not accept unverified apps, wait or use the official server.
- It reads GA4 only, not Universal Analytics, and never changes anything in your Analytics.
- It has no funnel report; the official server does.
- Analytics data passes through and is not stored, sold or used to train models.
- Free plan: 100 calls a month, summed across the connectors, no card and no time limit. Plans: https://myagents.click/en/products/connectors#pricing
Connector page: https://google-analytics.mcp.myagents.click/en
Top comments (0)