Best MCP for Google Analytics in 2026

Quick comparison: 9 Google Analytics MCP servers

A Google Analytics MCP server lets Claude, ChatGPT or Cursor run GA4 reports from a chat. The servers differ in who hosts them, whether they can change GA4 and what they cost. Every fact comes from the vendor's own docs or pricing page, checked November 7, 2026.

Hosted servers take a sign-in. Self-hosted servers need credentials and a machine that runs them. Composio and Zapier are general platforms where GA4 is one toolkit among many.

What is the best MCP for Google Analytics?

For most marketing teams, Ryze AI is our first pick: a hosted server with 9 read-only GA4 tools on the same connector as Google Ads, Meta Ads and Search Console. Disclosure: Ryze AI publishes this page and runs that server, so each reason comes with its limit.

Setup is a URL and a Google sign-in. No Google Cloud project, OAuth client or service account; the steps are on the Google Analytics MCP setup page. Limit: the tools need an active Ryze AI plan or free trial.

It covers the core reports. Standard, pivot and realtime reports, plus Admin API reads of key events, custom dimensions and data streams. Limit: no funnel report tool, which Google's official server has.

AI assistant traffic is built in. Four tools report AI-referred traffic from ChatGPT, Perplexity and other AI assistants by source, landing page, engine (against a prior period) and day; the source report adds conversions.

Ads and analytics in one chat. The connector also carries Google Ads (read tools and approval-gated write tools), Meta Ads, Search Console, TikTok and Shopify, 150+ tools in total. Limit: on GA4 it only reads, so it cannot create key events or custom dimensions.

Free and first-party with Python? Google's official server. Reporting across many sources on a small budget? Windsor.ai. Building your own agent? Composio, with 67 GA4 tools.

Side-by-side: access, setup and tool counts

Read means the server only pulls data; write means it can change GA4 settings or send data into GA4. Tool counts are the ones each vendor publishes. None of these servers was connected for this page.

Of the editions compared here, only Google's official server, Composio and Zapier can write to GA4 (CData sells a separate paid local MCP server that its README says can write). Google's keeps writes off by default. Ryze AI and surendranb's server are the two here that document ready-made reports or playbooks for AI assistant traffic.

The best Google Analytics MCP servers, ranked 1 to 5

1. Ryze AI

• What it does: hosted MCP server for Claude, ChatGPT, Cursor and Claude Code, with GA4, Google Ads, Meta Ads and Search Console on one connector.
• GA4 access: read only, 9 tools per the developers page: listAccountSummaries, runRawReport, runPivotReport, runRealtimeReport, runAdminRead (GET only) and 4 AI-referral reports.
• Setup: a connector URL and Google sign-in; Viewer access on the property is enough, per the Ryze AI help center.
• Price: needs an active plan or trial. Paid Ads is $89 a month flat with a 7-day free trial; annual billing saves 30% the first year (pricing).
• Fits: marketers who want GA4 next to ad spend. Limit: no GA4 writes, no funnel tool.

2. Google Analytics MCP (official)

• What it does: Google's open-source server for the Admin API and Data API, Apache 2.0, 3,675 GitHub stars on November 7, 2026.
• GA4 access: 13 tools in the README: 10 reads, including run_funnel_report and run_conversions_report (ad costs, ROAS, attribution models), and 3 writes (annotations, custom dimensions, key events) that need --allow-write and the analytics.edit scope. Google's developer guide still lists 9 read-only tools.
• Setup: pipx, a Cloud project with both APIs enabled, and Application Default Credentials. Google documents Claude Desktop, Claude Code, Cursor and Antigravity; it can also run over HTTP on Cloud Run. Google also runs a remote Data API endpoint with 4 read-only tools, not yet in its docs (Windsor.ai's GA4 MCP page).
• Price: free, apart from setup time and a machine to run it.
• Fits: developers who want free, first-party access. Limit: GA4 only, terminal setup.

3. Windsor.ai

• What it does: hosted MCP on Windsor's data platform, where GA4 joins 350+ other sources.
• GA4 access: read-only, per its GA4 MCP page; 92 GA4 metrics and 390 dimensions. No tool count published.
• Setup: add Windsor from Claude's Connectors, or its ChatGPT or Claude Code plugin, then connect GA4 once.
• Price: free plan with 1 source after a 30-day trial; Basic $23 a month ($19 billed annually); MCP on every plan (pricing).
• Fits: cross-channel reporting on a small budget. Limit: no tool list to review.

4. Adzviser

• What it does: hosted reporting connector in Claude's connector directory.
• GA4 access: read-only; its GA4 page lists 75 metrics and 161 breakdowns. No tool count published.
• Setup: open its directory listing in Claude, click Connect, sign in. No URL to paste.
• Price: Starter $48.99 a month for 3 sources, $0.99 14-day trial, unlimited MCP queries (pricing).
• Fits: the lightest install for a team that only reads. Limit: price grows per source.

5. Supermetrics MCP

• What it does: MCP endpoint over Supermetrics' marketing sources, GA4 included.
• GA4 access: the product page describes querying only and lists no tools.
• Setup: log in to the Supermetrics hub, add the endpoint, authorize with OAuth 2.0.
• Price: 14-day free trial. The pricing page lists Starter at €49 a month. The MCP page says any subscription includes MCP, while its FAQ ties it to plans with Data API access.
• Fits: existing Supermetrics customers. Limit: which plan includes MCP is unclear.

Ranked 6 to 9: developer and automation options

6. Composio

• What it does: developer platform whose SDK returns an MCP URL for its Google Analytics toolkit.
• GA4 access: 67 tools on the toolkit page: standard, pivot, funnel and realtime reports, plus writes such as Create Custom Dimension, Update Property and Send Events.
• Setup: managed OAuth; guides for Claude Code, Cursor and ChatGPT.
• Price: free Hobby plan with 100,000 tool calls a month (up to 20,000 through managed apps such as Google Analytics); Pro $29 a month (pricing).
• Fits: developers building an agent product. Limit: no built-in approval step; its FAQ advises adding one in the prompt.

7. CData

• What it does: an open-source local server (MIT) that reads Google Analytics as SQL tables, and Connect AI, CData's hosted MCP platform.
• GA4 access: the open-source server has 3 read-only tools: get_tables, get_columns and run_query.
• Setup: install and license the CData JDBC driver (a trial key works), build with Maven, add it to Claude Desktop.
• Price: Connect AI Standard $99 a month for 1 source, 14-day trial (pricing).
• Fits: data teams who prefer SQL. Limit: Java setup and a paid driver after the trial.

8. surendranb/google-analytics-mcp

• What it does: community GA4 server, MIT, 244 GitHub stars on November 7, 2026.
• GA4 access: 8 read tools in the README, including get_ga4_data and run_realtime_report, plus playbooks such as traffic diagnosis and AI referral analysis.
• Setup: uvx, npx or pip with a service account JSON file. Anonymous telemetry is on unless you set DO_NOT_TRACK=1.
• Price: free.
• Fits: developers who want analysis playbooks. Limit: service account setup.

9. Zapier MCP

• What it does: Zapier's MCP server, with Google Analytics 4 as one app.
• GA4 access: 6 actions on its GA4 MCP page: Run Report for a Property, Find Conversion, Create Conversion Event for a Property, Create Measurement and Send Measurement Events for an Application, plus API Request, a raw HTTP request that uses the GA4 sign-in.
• Setup: a Zapier account; pick the actions in the Zapier MCP dashboard.
• Price: 2 tasks from your Zapier plan per successful tool call (usage docs).
• Fits: one GA4 step in a wider automation. Limit: few ready-made report actions.

What Google Analytics MCP servers cost

Two servers are free and open source. Hosted ones charge per plan, per data source or per tool call. Entry prices as listed on November 7, 2026:

One cost sits outside the table: the AI client. Anthropic says custom connectors work on every Claude plan, but Free users are limited to one custom connector.

Read-only or write: what a GA4 MCP server can change

Most GA4 questions only need reads. Writes change how the property collects and labels data, so they need more care than a report.

What writes can do. Google's official server can add annotations, create custom dimensions and mark key events. Composio adds custom metrics, property updates and Measurement Protocol events. Zapier can create conversion events and send measurement events.

Scopes and roles. Google's server reads with the analytics.readonly scope; writes need analytics.edit plus the Editor role, or Marketer for key events. Ryze AI needs Viewer access, and its Admin API tool rejects create, update and delete paths.

A safe default. Start every server read-only. Turn writes on for one named task, check the result in GA4, then turn them off. For ads, keep write tools on approval; the Google Ads MCP holds each change until you confirm it.

How to choose a Google Analytics MCP server

Three questions settle it: do you need GA4 next to ad data, can someone run Python and a Cloud project, and does anything need to write to GA4?

Marketing team, GA4 next to ad spend

Ryze AI: GA4, Google Ads, Meta Ads and Search Console share one connector. Windsor.ai or Adzviser if you only report. For search data, see the Search Console MCP page.

Developer or analyst, free and first-party

Google's official server. surendranb's if you want playbooks built in, CData if your team prefers SQL.

Building an agent, or many sources

Composio for 67 tools behind one SDK, with your own approval step before writes. Windsor.ai, Supermetrics or Adzviser for many sources; compare per-source pricing first.

One GA4 step in an automation

Zapier MCP, if 2 tasks per call fits your plan. Agencies should also read how to connect multiple Google Ads accounts to Claude, and stores the best MCP for Shopify.

Prompts to test a Google Analytics MCP server

Run the same prompts on every server you shortlist. Replace the brackets with your own property and dates; the numbers that come back will be your own.

Returns accounts and properties (listAccountSummaries on Ryze AI, get_account_summaries on Google's server).

Returns one row per channel with both periods. Named date ranges keep the periods from being swapped.

Returns a ranked table of pages. Ask for keyEvents and sessionKeyEventRate; conversions and sessionConversionRate are deprecated.

Ryze AI answers with its AI-referral tools; elsewhere the assistant filters by session source, so check its list of assistants. More prompts: Claude skills for Google Analytics.

How we compared them

Sources

The verdict

Hosted servers take a sign-in; Google's official server is free and first-party but needs a Cloud project; Composio, CData and Zapier suit builders and automations. Keep GA4 writes off unless a task needs one.

For GA4 next to your ad accounts in one chat, Ryze AI is our pick; its 9 GA4 tools only read, so key events and custom dimensions stay a job for the GA4 interface.