Connecting GA4 to Claude or ChatGPT

The four options at a glance

Option 1: export and paste

Download a report or exploration as CSV and upload it to the chat. It works today with no setup, and it's fine for a one-off question.

The limits show up quickly: you have to know which report to export before asking the question, every follow-up needs another export, and the AI can't check anything you didn't include. It's also easy to paste a sampled or thresholded table without realizing it.

Option 2: Google's GA4 MCP server

Google publishes an experimental, open-source MCP server for Google Analytics that uses the GA4 Admin and Data APIs. You run it on your own machine with your own Google Cloud credentials, and it's read-only.

Good fit: an analyst or developer who wants free, official, local access and is comfortable with Google Cloud setup. Less good: a marketing team where several people need it, or anyone using Claude or ChatGPT on the web, which need a remote server rather than a local one. Install from Google Analytics' own GitHub organization; there are several similarly named community projects.

Option 3: a hosted MCP connector

A hosted connector runs the MCP server for you. You add one URL to Claude or ChatGPT, sign in with Google in the browser, choose which properties the AI can see, and ask questions. Every teammate signs in as themselves.

1. Create a MegaMCP workspace.

   Start a free trial; no credit card is needed.

2. Connect Google.

   In Connections, sign in with the Google account that has access to your GA4 properties. Switch off any property the AI shouldn't see.

3. Add MegaMCP to your AI app.

   Paste https://megamcp.com/api/mcp as a custom connector in Claude or a custom app in ChatGPT, then sign in.

4. Ask a question you know the answer to.

   Compare last week's sessions with the GA4 interface to confirm the property, date range and time zone match.

MegaMCP's GA4 tools cover flexible Data API reports, period comparisons that explain what drove a change, top pages and landing pages by source/medium, funnels, pivot tables, realtime, audience breakdowns, a property setup audit and change history. Admin changes (key events, data retention, enhanced measurement, data streams, custom definitions) are proposals with before and after values that a person approves, and plans that delete a stream or shorten retention say plainly that they can't be undone. See the GA4 integration.

Option 4: BigQuery export

GA4 can export raw event data to BigQuery daily (and streaming, at a cost). An AI with a BigQuery tool can then write SQL over every event, with no sampling or thresholds.

It's the most powerful option and the most work: you need the export set up in advance (it doesn't backfill history), you need to understand the event schema, and queries cost money. MegaMCP's BigQuery integration shows a dry-run byte estimate before you approve any query; it's marked Rolling out.

GA4 quirks that trip up AI answers

• Processing delay. Google says GA4 data can take 24 to 48 hours to process fully. Yesterday's numbers may still rise. Ask for complete periods ending two days ago when precision matters.
• Key events, not conversions. Since 2024 GA4 calls conversions “key events”; Google Ads conversions are configured separately. Make sure the AI uses the key event you care about, not every event.
• Thresholds. When Google signals or small numbers are involved, GA4 may withhold rows to protect privacy, so totals by dimension don't always add up.
• Time zones and currencies. Reports follow the property's settings. Compare like with like.
• Attribution differences. GA4 and ad platforms credit conversions differently. Expect GA4 paid conversions to differ from what Google Ads or Meta report.
• UTM hygiene. Channel reports are only as good as your UTM parameters. Inconsistent casing splits sources.

Good first questions

Start with questions that combine dimensions you'd normally build an exploration for:

• Reads data

  Why did key events drop last week compared with the week before? Break it down by default channel group and landing page.

  Tools it calls: ga4_compare_periodsga4_landing_pages

• Reads data

  Which landing pages get over 1,000 sessions a month but convert under 1%?

  Tools it calls: ga4_landing_pages

• Reads data

  Audit this property's setup: time zone, currency, data retention, enhanced measurement and key events.

  Tools it calls: ga4_property_audit

The GA4 prompt library has 20 more, grouped by reporting, optimization, audits and agency work.

Example prompts

Copy these into Claude, ChatGPT or Cursor with MegaMCP connected. Each lists the tools it typically calls; more are in the prompt library.

• Reads data

  Give me a weekly GA4 traffic summary I can paste into Slack, compared with the previous week.

  Tools it calls: ga4_traffic_summary

• Reads data

  How many people are on the site right now, and from which countries?

  Tools it calls: ga4_realtime

• Reads data

  Run a funnel from product view to add to cart to purchase for the last 30 days.

  Tools it calls: ga4_funnel_report

• Reads data

  Which settings changed in this GA4 property in the last month, and who changed them?

  Tools it calls: ga4_change_history

• Proposes a change

  Mark the generate_lead event as a key event and show me the plan.

  Tools it calls: ga4_list_key_eventsga4_create_key_event

  Any change becomes a plan you review and approve. Nothing moves until you do.