Google Marketing Stack MCP

BuildsClose

Ask Claude about your GA4, Search Console, Tag Manager, PageSpeed and Sheets data, in plain English.

LiveOpen sourceMIT

GitHub npm package

GA4
15 tools

Search Console
9 tools

Tag Manager
8 tools

PageSpeed
5 tools

Sheets
5 tools

Account
2 tools

44 tools in 6 groups. Hover a square to read its name.

What it is

A Model Context Protocol (MCP) server that wraps five Google APIs into 44 tools an AI assistant can call. You connect it once, then ask questions such as "which queries rank between position 4 and 10?" and get real answers from your own accounts.

It runs on your machine and uses your own Google login, so your data never touches a third-party server. It works with Claude Code, Claude Desktop and Cursor, and it is published on npm, so it runs without cloning. It has 7 GitHub stars and 450+ npm downloads in the last 30 days, and marketers in San Francisco (SF) use it.

The problem

Marketing data lives across many dashboards. Getting an answer meant exporting CSVs, clicking through reports, writing custom queries or asking a developer to pull it. The server replaces all of that with one question.

How it works

Create Google credentialsIn a Google Cloud project, enable the Analytics Data, Search Console, Sheets and Tag Manager APIs, then create an OAuth client of type Desktop app.
Log in onceStart the server with the client ID and secret. A browser window asks you to approve access, and the token is saved locally for every later run.
Connect your AI clientAdd the server to Claude Code, Claude Desktop, Cursor with your GA4 property, Search Console site and GTM IDs.
Ask in plain EnglishThe assistant picks the right tool for the question, for example ga4_landing_pages or gsc_top_queries, calls it, and answers from the real data.

Architecture

Legend

MCP tool call
Local token read

Use any one

Claude Code

Claude Desktop

Cursor

Server

MCP server

Node.js, stdio
44 typed tools

Token on your machineLog in once

Google APIs

Google Analytics 415 tools

Search Console9 tools

Tag Manager8 tools

PageSpeed5 tools

Sheets5 tools

Credentials stay on your machine. Only the server talks to Google.

The other 2 of the 44 tools are get_status and reauthenticate. They have no card.

Impact

44tools across 6 groups

7GitHub stars

450+npm downloads in the last 30 days

SFmarketers in San Francisco use it

2of my other builds run on it

Where the 44 tools go

Tool count and access per group, from the repository.

GA4

15Read

Search Console

9Read

Tag Manager

8Read

PageSpeed

5Read

Sheets

5Read and write

Account

2Sign-in

It is the Google layer underneath the Growth Intelligence Stack and the SaaS SEO Agent, so both depend on it.

Built with

Node.jsModel Context ProtocolGoogle OAuth 2.0GA4 Data APISearch Console APISheets APITag Manager APIPageSpeed Insights

Decisions that mattered

Desktop-app OAuth, not Web. A web client needs pre-registered redirect URIs and fails on the random local port the server uses for the callback.
Runs locally. Nothing is stored or proxied, so there is no service to secure and no data to leak.
All four APIs are enabled up front, because the login requests every scope at once.
Published on npm, so one npx command starts it without a clone.

What is next

More tools as new questions come up.

Sources: Repository README and docs; GitHub and npm public stats, read 7 Nov 2026