Can you just connect Shopify, Meta and GA4 to Claude yourself?

Yes. Shopify, Meta and Google each have a free, official way into Claude. Each answers its own questions well. None of them follows an ad click to the visit and the order.

Raphael Gonzalez, Founder, Anry
SourceFree, official route into ClaudeAnswers wellCan't do
ShopifyShopify's connector in Claude's directoryOrders, products, customers, salesSee ad spend or what visitors do on the page
MetaMeta's Ads MCP, mcp.facebook.com/adsSpend, cost per result, delivery, pixel healthSee your real orders
GA4Google's Analytics MCP, run on your computerSessions, sources, funnels, realtimeMake changes, or count visitors who block tracking
All three in one chatYes, switch all three onEach platform's own questions, side by sideJoin one ad click to its visit and its order

Yes, and it's a good place to start

Every piece is free and made by the platform itself. I'd set it up before paying for anything, including us.

  • Shopify. Add the connector from Claude's directory and approve it in your Shopify admin. It reads orders, customers and sales, and can edit products if you let it. Steps in How to connect Claude to Shopify.
  • Meta. Add https://mcp.facebook.com/ads as a custom connector and pick your business portfolio. Reporting, delivery errors, pixel health, and paused-by-default ad creation. Steps in the Meta Ads MCP setup.
  • GA4. This is the fiddly one. Google's server is open source and labelled "Experimental". It runs on your own machine through pipx, after you enable the Analytics Admin and Data APIs in a Google Cloud project and log in with gcloud on the analytics.readonly scope (GitHub). That means Claude Code, Cursor or Claude Desktop's config file. There's no Google-hosted version to paste into claude.ai: Analytics isn't on Google's list of remote MCP servers (Google Cloud). It has seven tools, all read-only, including run_report and run_funnel_report.

What the stack answers well

Anything that lives inside one platform, asked in plain words:

  • "What were sales, orders and AOV last month against the month before?" (Shopify)
  • "Which campaigns spent the most this week, and at what cost per purchase?" (Meta)
  • "Are any ads or my catalog blocked from delivering?" (Meta)
  • "Sessions and conversion rate by source and device for the last 28 days." (GA4)
  • "Where does the checkout funnel drop, step by step?" (GA4)

You can also put two answers next to each other. "Meta spend by week, next to Shopify revenue by week" works fine. For a weekly check-in on a store with one or two channels, that's a lot of value for no money.

Where it falls short

1. Nothing joins the ad click, the visit and the order

Each server answers from its own records. Meta knows the click. GA4 knows the session. Shopify knows the order. None of the three hands the model a row that says this ad's click became this visit became this order. So when you ask "which ad brings buyers who actually pay?", Claude lines up three separate reports by date or by UTM tag and guesses at the overlap.

We hit this on anry.io. Meta sold us landing page views that looked cheap, and only joining each click to its visit showed that almost none of those people did anything once the page loaded (the campaign, in full). Two reports side by side wouldn't have shown it.

2. The three numbers won't agree

They're not supposed to. Meta credits purchases inside its attribution windows and models part of iOS traffic (Meta). GA4 only counts visitors with cookies and JavaScript on, misses people with ad blockers, and may use another time zone than Shopify (Shopify Help). GA4 also withholds rows when Google signals is on and counts are low, and you can't switch that off (Google). Shopify counts the orders.

Put all three in one chat and you have three answers to the same question, unless you tell Claude which one counts. Decide that before you trust a combined answer. A sensible split: Shopify for revenue, the ad platform for spend. If the gap itself is what worries you, why ROAS drops when nothing changed goes deeper.

3. The model does the maths on raw rows

Every total in a combined answer is Claude adding up rows it pulled into the chat. Shopify's connector lists orders 50 at a time, for example (Shopify Help), so a month of orders is several calls, and a missed page means a wrong total that looks right. Ask it to show each query and its row count, and check one figure against the platform's own dashboard before you act on it.

4. It takes setup, and it takes room

GA4 alone is a Google Cloud project, two APIs and a command-line login. Then there's the tool list. On my setup Meta's server loads 97 tools, Shopify's connector has 25 (Claude) and GA4 has 7. That's about 129 tool descriptions in every chat before you've asked anything. Long tool lists crowd the context, which is why Klaviyo, for one, ships a switch to load only its core tools (Klaviyo docs). Switch on only the connectors a question needs.

5. It only answers what you ask

These servers respond to questions. None of them watches the store between chats, so a page that stopped selling on Tuesday stays quiet until you happen to ask about that page.

When doing it yourself is enough

  • You want reporting in plain words: sales, spend, sessions, one platform at a time.
  • You run one channel, or your ad spend is small enough that a wrong guess doesn't cost much.
  • You're happy to check the joins yourself, with UTM tags set up properly on every ad.
  • You want Claude to make changes: draft Meta ads, edit Shopify products. That's the official servers' job anyway.

Paid bundle connectors save the setup. Porter Metrics, for example, says its connector covers Meta, GA4, Shopify and more behind one URL (Porter). Before you pay, check whether they join data at the visit level or put each platform's report side by side.

Where Anry fits

Anry is the join. Each Meta or Google ad click becomes one line that runs from the visit to the order, so it can point to the page, device or ad where sales slip away and attach a price taken from your store's orders. It's a read-only MCP server that sits next to the free ones in Claude, Claude Code or Cursor, and the official servers stay in charge of making changes. If you spend little on ads, or you mainly want reports, the free stack above is enough.

Frequently asked questions

Can Claude analyze Shopify, Meta Ads and GA4 data together?

Yes, if you switch all three connectors on in one chat. It can put their reports side by side. It can't follow a single ad click to the order, because no one of the three servers holds that link.

Is there an official GA4 MCP server?

Yes. Google Analytics publishes an open-source, experimental server on GitHub. It's read-only, runs on your own computer with Python and Google Cloud credentials, and has seven reporting tools.

Can I add the GA4 MCP to claude.ai?

Not directly. Claude's custom connectors take a remote URL, and Google's Analytics server runs locally. Use Claude Code, Cursor or Claude Desktop's config file, or a third-party remote server.

Why don't Meta, GA4 and Shopify show the same numbers?

Meta credits conversions in its own attribution windows and models some of them. GA4 misses visitors who block cookies or tracking. Shopify counts real orders. Pick one source per number and stick to it.

Are the official MCP servers free?

Shopify's connector has no extra Shopify fee, Meta's Ads MCP is free during its open beta, and Google's GA4 server is open source. You still pay for your Claude plan.

Sources

See which ad clicks leave your store before they buy.

14-day free trial, no card. Works in Claude, Claude Code and Cursor.

Start free trial