Build1 publisher3 min readPublished
Google's Analytics MCP repo predates MCP's donation to the Linux Foundation
Google's server for Google Analytics is a read-only Python package you run locally through pipx, and it has been on GitHub since mid-2025. Anthropic only handed the protocol to the Linux Foundation that December.
The Engineer · Build desk
What happened
- The googleanalytics/google-analytics-mcp repo is an MCP server Google wrote itself and publishes under its Google Analytics GitHub organisation, letting an AI client query GA data with no CSV export first.
- Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation on 9 December 2025, with Block and OpenAI as co-founders and Google, Microsoft, AWS, Cloudflare and Bloomberg as supporters.
- Access runs through Application Default Credentials on the analytics.readonly scope, so the server reads properties the signed-in account can already see and cannot modify any data.
- The package reports 73,067 downloads in the last month, 12,455 in the last week and 2,071 in the last day, a monthly figure about 23 times the repository's star count.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- contradiction The post reads Google's server as a consequence of neutral governance, but the creation date puts the code first, so an integration plan premised on the foundation unlocking first-party servers is built on the wrong causality.
- cost Adoption is per operator: every machine that runs the server needs Python 3.10 or newer, pipx, and its own gcloud login by an account that already holds access to the property.
- decision Shipping this into client work means accepting an Experimental tag in a deliverable and verifying behaviour against the code, because the published documentation describes an older release.
- capability run_funnel_report puts drop-off questions in the chat instead of the GA console, so someone can ask which step loses users without building the funnel report by hand first.
Registration differs by client. Claude Code takes one line: `claude mcp add analytics-mcp --scope user`, two `-e` variables for GOOGLE_APPLICATION_CREDENTIALS and GOOGLE_PROJECT_ID, then `-- pipx run analytics-mcp` [15]. For Gemini CLI the README's path is editing `~/.gemini/settings.json` and adding an entry under `mcpServers` [16]. Google documents two ways to put the credentials in place, both requesting the `analytics.readonly` and `cloud-platform` scopes: an interactive `gcloud auth application-default login` with `--client-id-file`, or the same command with `--impersonate-service-account` [14].
The dev.to post by Nokka that assembled these figures argues Google went ahead and built its own server because MCP stopped belonging to one company on 9 December 2025 [6]. The repo's own timeline complicates that. It was created in mid-2025 [17], roughly six months before the donation [4]. What did move in that window was the protocol's install base: about 100,000 downloads in the SDK's first month after the November 2024 launch, and 97 million a month by March 2026 [3][4], a factor of 970 [5].
Divide the monthly downloads by 23 and the star count lands near 3,180 [1]. The post attributes the gap to the install path, since `pipx run` means nobody clones the repo and people who never open it do not star it [9]. The daily figure times 30 gives 62,130, about 85 percent of the monthly total [2], so the three windows overlap. For 73,067 to stand in for 73,000 users, each machine would have to fetch the package about once a month, and no CI job could fetch it at all.
Google's documentation page for the server was last updated 16 September 2025 while the code was last updated in August 2026 [18], a gap of about eleven months [3]. The post recommends anchoring on the repo README for that reason [23]. The Experimental tag has now stood through more than a year of installs [6].
The tools are grouped by task type into three sets [10]. `run_funnel_report` is the one that moves work out of the console, because funnels normally have to be configured in the GA interface itself [10]. Google's own examples show the range of what the server is expected to carry: one asks how many users there were yesterday, another asks for a data-driven marketing plan against a 5,000 dollar monthly marketing budget, which requires the model to pull several datasets and join them without help [19]. The README's examples are narrower, such as which events were most popular in a property over the last 180 days, and whether most users were logged in over the past six months [20].
Every number here comes from one post, published 15 September 2026, whose byline states the draft was produced by a deepseek-v4.1-flash model through Hermes Agent from Nous Research and then reviewed and edited by Nokka [22]. The same post puts googleapis/mcp-toolbox at 16,405 stars and gcloud-mcp at 903 [21].
What to watch
- Whether the Experimental label comes off the repo, and whether Google publishes a hosted variant alongside the local pipx path.
- Whether Google's documentation page for the server is updated past 16 September 2025 to match the August 2026 code.
- Whether any tool ships that needs a scope beyond analytics.readonly, which would change the read-only risk profile.