A GA4 MCP server lets Claude or ChatGPT query Google Analytics 4 and change its setup from a chat. PaidSync.ai is a hosted GA4 MCP server: 26 GA4 tools, 13 of them writes, that report from the Data API and Realtime API and create properties, data streams, conversion events (key events), custom dimensions and metrics, audiences and Google Ads links, in the same connection as Google Ads, Tag Manager and Search Console.
To use it, in order: create a PaidSync.ai account at paidsync.ai; connect Google Analytics 4 in the dashboard with the Google account that can see the property; add PaidSync.ai in Claude or ChatGPT and sign in (the client-by-client guides are at the end of this page); then ask "Which events should be marked as conversions but are not?" Reads answer at once; setup changes apply on the first call, and deletes are confirmed first.
An MCP server turns an API into typed tools an AI assistant can call. Connect one to Claude or ChatGPT and "which channels drove the biggest week-over-week conversion drop, and are there any events that should be marked as conversions but are not" becomes something the assistant can check, propose, and, once you approve, execute against the property. This page covers PaidSync's GA4 surface in full: the 26 tools, the analysis-plus-configuration split, the checks on writes, and how it fits with the GTM and Google Ads MCPs in the same server.
It is written for both audiences. Analysts and marketers who want GA4 answers and configuration through chat, and agent builders who want to know the protocol surface before wiring it into a workflow. Each section is labeled.
What a GA4 MCP server does
Two things, in practice. It reads, so the assistant can query the GA4 Data API and Realtime API in natural language and pull metadata, audiences, events, and property configuration. It configures, so properties, data streams, conversion events, custom dimensions and metrics, audiences, Google Ads links, and enhanced measurement can be created, updated, or deleted through conversation. Configuration writes apply on the first call, and destructive operations require an explicit confirm step.
The bit that most GA4 MCP servers do not cover: end-to-end setup from an ad platform through GTM into GA4. Because PaidSync includes GTM and Google Ads in the same server, "add a lead-form-submit conversion, wire it in GTM, mark it as a conversion in GA4, and import it as a conversion in Google Ads" is one conversation, one step at a time.
The official GA4 MCP, compared honestly
Google's own README says "This repo contains the source code for running a local MCP server that interacts with APIs for Google Analytics." and heads the page "Google Analytics MCP Server (Experimental)" (github.com/googleanalytics/google-analytics-mcp, read 26 Oct 2026). PaidSync.ai is hosted and also writes the setup. Google publishes an official Google Analytics MCP server at github.com/googleanalytics/google-analytics-mcp. It is labelled experimental, it is read only, and it asks for the analytics.readonly scope. You run it on your own machine through pipx, and the README documents Gemini CLI, Gemini Code Assist and Claude Code as clients. Seven tools cover account summaries, property details, Google Ads link listing, standard, funnel and realtime reports, and a custom dimensions and metrics listing. Checked September 2026.
That makes it a good first-party read path and nothing more. PaidSync's GA4 coverage sits at the configure tier: 26 tools, so the assistant can create and delete key events (conversion events), custom dimensions and metrics, audiences and Google Ads links. Each write applies on the first call, with an explicit confirm on deletes. Where the official server is the better pick, the table says so.
| Dimension | PaidSync | Google official GA4 MCP |
|---|---|---|
| Who hosts it | PaidSync, one endpoint at mcp.paidsync.ai/mcp | Google, open source, runs locally through pipx. Labelled experimental |
| Read or write | Configure tier, 26 tools. Key events, custom dimensions and metrics, audiences, Google Ads links, data streams | Read only, scope analytics.readonly |
| Which clients | Claude, Claude Code, ChatGPT, Cursor, Copilot Studio and any MCP client over Streamable HTTP | Gemini CLI, Gemini Code Assist and Claude Code documented in the README |
| Platforms covered | 14 platforms, GA4 next to Google Ads, GTM, Meta and more in one session | GA4 only |
| Approval gating | confirm_destructive on delete calls, approval cards in the PaidSync workspace; no dry_run on GA4 writes | Not applicable, nothing writes |
| Tools | 26 | 7 listed in the README |
| What PaidSync adds | Configuration writes, the GA4 Auditor agent, GA4 and GTM in the same prompt, approval cards with Undo where a safe inverse exists, and a best-effort 7-day before-and-after log | First party, free, canonical read path |
Quick verdict.
You want a free, first-party, read-only GA4 query path in Gemini CLI or Claude Code, and running it locally is fine: Google's official server.
You need to create a key event, add a custom dimension, link Google Ads, or audit GA4 next to GTM and the ad platforms from one chat: PaidSync. The writes exist, and each one waits for your tap.
You want both: run both. The official server for reads in a local IDE, PaidSync for configuration and the cross-platform view.
All 26 GA4 tools by category
PaidSync's live server exposes 26 GA4 tools (server total 630+ across 14 platforms). The table names the operational core; the full current catalog is discoverable live from any session via paidsync_context with include_tools:['ga4'].
| Category | Named tools |
|---|---|
| Account and property | connect_ga4, ga4_auth_status, list_ga4_accounts, list_ga4_properties, get_ga4_property, create_ga4_property |
| Data streams | list_ga4_data_streams, create_ga4_data_stream |
| Conversion events | list_ga4_conversion_events, create_ga4_conversion_event, delete_ga4_conversion_event |
| Custom dimensions and metrics | list_ga4_custom_dimensions, create_ga4_custom_dimension, list_ga4_custom_metrics, create_ga4_custom_metric |
| Audiences | list_ga4_audiences, create_ga4_audience |
| Google Ads links | list_ga4_google_ads_links, create_ga4_google_ads_link, delete_ga4_google_ads_link |
| Enhanced measurement | get_ga4_enhanced_measurement, update_ga4_enhanced_measurement |
| Analysis | run_ga4_report, run_ga4_realtime_report, get_ga4_metadata |
| End-to-end setup | setup_ga4_conversion_tracking |
Four of these carry most of the operational work. run_ga4_report answers arbitrary Data API questions (metrics, dimensions, filters, date ranges) so cross-cuts that take multiple clicks in the GA4 Explore module take one prompt here. create_ga4_conversion_event plus setup_ga4_conversion_tracking mark events as conversions and, when GTM is connected, sequence the tag creation on the GTM side. create_ga4_google_ads_link links a GA4 property to a Google Ads account so the conversions actually flow into Google Ads for bidding. And create_ga4_audience builds an audience that can be exported to Google Ads for remarketing without leaving the conversation.
For marketersGA4 answers and setup in one conversation
The practical shape of a week on PaidSync's GA4 stack looks like this. Monday morning you ask "which channels lost conversions week over week, and is the drop concentrated on one device or geography". The assistant runs run_ga4_report against the Data API and answers in plain numbers. You spot a checkout event that should be a conversion but is not; ask for a fix. The assistant proposes create_ga4_conversion_event, you confirm, and it lands.
The same session pairs with the GTM MCP for the tag half. Ask "add a purchase conversion" and the assistant sequences the GTM tag creation, publishes the workspace, marks the GA4 event as a conversion, and creates the GA4 Google Ads link if it does not yet exist. Each step is its own tool call, so you can stop between them. Because Google Ads is in the same server, the follow-up "import that conversion in Google Ads and switch the campaign bidding to tROAS" runs in the same conversation. That composite is the practical difference between a GA4 connector and a measurement server.
Trust is procedural, not promissory. GA4 writes apply on the first call, so the check is your assistant's confirmation prompt over MCP, or an approval card in the PaidSync workspace. Deletes require a second explicit confirmation.
Connect a GA4 property to Claude, ChatGPT, or Gemini and run the first weekly report in minutes.
Start free, 100 tasks, no card See the GTM half of the setupThe protocol surface behind the 26 tools
PaidSync is one MCP endpoint covering 14 platforms; details on the MCP server page. To keep client context small, tools/list returns a lean starter set. The full catalog is discovered at runtime through paidsync_context, which returns the active account, connected platforms, and the grouped tool catalog (include_tools:['ga4'] returns the exact GA4 tool names), and executed through paidsync_exec. That two-call pattern means an agent can enumerate and invoke all 26 GA4 tools without a 630+-tool schema payload sitting in every conversation.
The server-side checks run whatever the client. GA4 writes have no dry_run and apply on the first call, and destructive operations (delete conversion event, delete Google Ads link) require confirm_destructive. For discovery and verification, the domain publishes OAuth authorization-server and protected-resource metadata (RFC 8414 and RFC 9728), an MCP server card, agent-auth.json, security.txt, and a Content-Signal directive.
PaidSync vs community GA4 MCP servers vs Zapier
Two other paths to GA4 through MCP show up on current lists. Open-source community GA4 MCP servers on GitHub typically wrap the GA4 Data API for querying; self-hosted, single-account, and read-focused. Configuration writes are not the primary use case. Zapier's Google Analytics MCP surfaces Zapier's existing Google Analytics integration through the MCP protocol; useful for lifecycle triggers (new session, goal completion) piped to other apps, less useful for configuration work. PaidSync is hosted, covers both analysis and configuration, and sits inside a server that also runs Google Ads, Meta, LinkedIn, TikTok, and GTM in the same session.
| Dimension | PaidSync | Community GA4 MCP | Zapier Google Analytics MCP |
|---|---|---|---|
| GA4 tools | 26 named, analysis + configuration | Read-focused subset | Lifecycle triggers, not configuration |
| Data API and Realtime API | Yes, run_ga4_report, run_ga4_realtime_report | Typically yes | Not the model |
| Create conversion events | Yes, create_ga4_conversion_event | Rare | No |
| Custom dimensions and metrics | Yes | Rare | No |
| Google Ads link management | Yes, create and delete | No | No |
| Enhanced measurement | Yes, get and update | No | No |
| Pairs with GTM in the same session | Yes, /gtm-mcp | Separate server if any | Zapier apps, not MCP-native GTM tools |
| Checks on writes | confirm_destructive on deletes; no dry_run on GA4 writes | Not applicable | Through Zapier auth prompts |
| Hosted vs self-hosted | Hosted | Self-hosted | Hosted |
| Price | Free $0 (100 tasks), Pro from $99/mo | Free, self-hosted | Free tier plus paid Zapier plans |
Quick verdict.
You want a free self-hosted GA4 MCP for personal Data API queries: a community GA4 MCP on GitHub.
You want GA4 lifecycle events piped into other apps and you already live in Zapier: Zapier's Google Analytics MCP.
You want to analyze GA4 and configure conversion events, custom dimensions, and Google Ads links from chat, and pair with GTM plus Google Ads in one session: PaidSync.
What this does not do
Stated plainly, because scoping a tool accurately beats overselling it.
- No Universal Analytics. UA sunset in 2023-2024 and PaidSync targets GA4 only.
- No BigQuery export management. BigQuery export is configured in GA4 admin today; querying the resulting dataset lives in BigQuery.
- No user or property permission management. Access control stays in GA4 admin.
- No dry run on GA4 writes, and no approval card over MCP. Each GA4 change applies on the first call. When your own assistant calls PaidSync over MCP, its confirmation prompt is the only human check, so keep it on for write tools.
Pricing
Every plan includes all 14 platforms and there is no per-property or per-platform charge. Every tool call counts as one task, whether it runs a report or creates a conversion event. Full detail at paidsync.ai/pricing.
| Plan | Price | Included |
|---|---|---|
| Free | $0/mo | 100 tasks a month, no card required |
| Pro | From $99/mo | 1,000 tasks ($99), 2,500 tasks ($149), or 6,000 tasks ($199) |
| Team | From $249/mo | 10,000 shared tasks, up to 20 seats |
| Done For You | Custom | Fully managed paid media and measurement |
Put 26 GA4 tools, plus GTM and Google Ads in the same session, in front of your AI assistant today.
Connect Google Analytics 4 to your AI
One setup per client, same endpoint and sign-in everywhere.
- Google Analytics 4 to Claude
- Google Analytics 4 to Claude Code
- Google Analytics 4 to ChatGPT (full guide)
- Google Analytics 4 to Gemini (full guide)
- Google Analytics 4 to Microsoft Copilot Studio
- Google Analytics 4 to Perplexity (full guide)
- Google Analytics 4 to Cursor
- Google Analytics 4 to Codex
Frequently asked questions
What is a GA4 MCP server?
An MCP (Model Context Protocol) server exposes GA4 API actions as typed tools that AI assistants like Claude, ChatGPT, and Gemini can call. A GA4 MCP lets the assistant query the GA4 Data API and Realtime API in natural language and configure the property (create conversion events, custom dimensions and metrics, audiences, and Google Ads links) through conversation instead of clicks in the GA4 admin. PaidSync's GA4 surface is 26 tools spanning both analysis and configuration.
Is there an official GA4 MCP server?
Yes. Google publishes an experimental GA4 MCP server that runs on your own machine and only reads (checked 29 Sep 2026). PaidSync has 26 GA4 tools. Most read your reports; the rest make changes you ask for, such as marking key events, creating custom dimensions or linking Google Ads.
How many GA4 tools does PaidSync have?
26 live GA4 tools, inside a server of 630+ tools across 14 platforms. The GA4 set covers account and property, data streams, conversion events, custom dimensions and metrics, audiences, Google Ads links, enhanced measurement, analysis (Data API + Realtime API), and end-to-end conversion tracking setup.
Can PaidSync set up GA4 conversion tracking from chat?
Yes, this is the headline use case. setup_ga4_conversion_tracking wires a conversion event end-to-end. Paired with the GTM MCP, the assistant can create the matching tag on the GTM side and publish the workspace. Paired with the Google Ads MCP, the same conversation imports the conversion into Google Ads for bidding.
How does PaidSync compare with community GA4 MCP servers?
Open-source GA4 MCP servers on GitHub typically wrap the GA4 Data API for querying; self-hosted, single-account, focused on reads. Configuration writes are not the primary use case. PaidSync is hosted, covers both analysis and configuration, and pairs with the GTM MCP in the same conversation for end-to-end conversion tracking work.
Is it safe to give an AI write access to a live GA4 property?
GA4 writes have no dry run and apply on the first call, so the human check matters. In the PaidSync workspace every write is an approval card by default. When your own assistant calls PaidSync over MCP, PaidSync writes no card, so the check is the assistant's own confirmation prompt. On both paths, destructive operations (delete conversion event, delete Google Ads link) require an explicit confirm_destructive step, and applied changes are logged, on a best-effort basis, with a 7-day before and after.
How much does the PaidSync GA4 MCP cost?
Free is $0 for 100 tasks a month, worth $50, with no card. Pro starts at $99 per month for 1,000 tasks, with 2,500 tasks at $149 and 6,000 at $199. Team starts at $249 per month for 10,000 shared tasks and up to 20 seats. Done For You is custom-priced fully managed paid media and measurement.
Which AI assistants work with the PaidSync GA4 MCP?
Claude (web, desktop, and Claude Code), ChatGPT, Gemini, Perplexity, Cursor, Windsurf, and any client that speaks MCP. Setup guides: Claude, Claude Code, ChatGPT, Gemini.