Google Analytics 4 MCP server for your AI

Yes. You connect with your own Google OAuth client, so the connect form asks for an OAuth Client ID and an OAuth Client Secret from that project. Enable both the Google Analytics Admin API and the Analytics Data API under APIs & Services -> Library first, then authorize; those two APIs are what the reporting and admin calls run against. You never paste an API key: the last step is a normal Google consent screen.

You authorize with the Google account that already has access to your GA4 properties, and after the consent screen you choose which of those properties to connect and which workspace each one belongs in. The agent works against the properties you picked, not everything the Google account can see. Reporting and admin calls run inside that scope.

Both, and you decide which. Most of the catalog is read-only reporting and admin lookups, and the writes are narrow: creating an audience, creating a custom dimension, marking an event name as a key event, and updating property fields such as display name, time zone or currency. Set the connection to read-only if you only want analysis, or keep full access with approval required so those four kinds of change wait for a human.

ga4_run_realtime_report covers the last 30 minutes of activity, which is what you want when someone asks whether the site is up or a campaign has landed. For historical work, ga4_batch_run_reports sends up to 5 reports in a single batch, and there is a pivot equivalent for breakdowns that need two dimensions crossed. Batching matters when an agent would otherwise fire one call per date range.

The GA4 Data API rejects dimension and metric combinations that cannot be queried together, and this is the usual reason a first attempt fails. ga4_check_compatibility tests a request before it runs, and ga4_get_metadata lists every dimension and metric available on the property, including the custom ones you have defined. Pointing the agent at those two first turns a guessing loop into one clean query.

The triggers are threshold-based and tuned to the questions analytics owners get paged about: sessions dropping below the same period last week, traffic spiking above its normal range, conversions or key events falling under their recent average, purchase revenue missing target, one channel such as organic or paid search dropping sharply, engagement rate falling, and live visitors crossing a level you set. Each fires against the property you connected, so your AI client can investigate with a report before you have even opened GA4.

No. PopMCP hosts it and gives you a Streamable HTTP MCP URL. There is nothing to deploy, no callback URL to register, and no infrastructure to keep running or patch.

Any client that supports a remote Streamable HTTP MCP server — Claude Desktop, Claude Web, ChatGPT custom connectors, Codex and the Codex CLI, OpenCode and Z.ai's ZCode, as well as editors such as Cursor, Windsurf and Zed. Your client signs in to PopMCP with OAuth; the provider credential never leaves the server.

Loading 21 tool definitions into every conversation burns context before the work starts. PopMCP loads a curated 21-tool set covering the everyday workflows and keeps the rest searchable, so your AI pulls a specialist tool only when a task needs it. You can switch a connection to the full catalog at any time.

A tool is something your AI calls — there are 21 here. A trigger is something PopMCP watches for you: 7 Google Analytics 4 events that can start work on the PopMCP side. Triggers never enter your AI client's context, and none of them ask you to paste a callback URL into Google Analytics 4.