Meta Ads MCP and Google Ads MCP: What Claude Gets Right, and Where the Numbers Go Wrong

Meta and Google now both offer an official MCP server, which means you can connect your ad accounts to Claude or ChatGPT and ask about campaigns in plain English. The Meta Ads MCP takes a few minutes to switch on; the Google Ads MCP takes an engineer and an afternoon. Both work, and both have the same catch: the AI reads each platform's own numbers, so the answers come back in the revenue each platform credits to itself, not in profit.

Here's what each one does, how to connect it, where the answers go wrong for an ecommerce brand, and how to get profit answers instead of ROAS answers.

What is the Meta Ads MCP?

The Meta Ads MCP is Meta's own MCP server for ad accounts, sold as Meta ads AI connectors. Meta opened it in beta on 29 April 2026, according to PPC Land, and its Business Help Center guide lists ChatGPT, Claude, Claude Code and Perplexity as supported agents, all using the same remote server at mcp.facebook.com/ads.

It does far more than reporting. Through the connector an agent can pull performance reports, create and edit campaigns, ad sets and ads (new ads start paused), manage custom audiences and catalogues, check signal health, run A/B tests and lift studies, and read the account's activity log. Meta notes that not every advertiser has access yet, and that connecting an account through its MCP server won't on its own get the account banned.

How to connect Meta Ads to Claude or ChatGPT

1. Set the guardrails first. In Meta Business Suite go to Settings, then Integrations, then Ads MCP server, pick the ad account and block anything you don't want an agent doing, such as creating campaigns or raising budgets above a set amount.
2. Add the connector in your AI tool. In Claude or ChatGPT, add Meta's ads connector from the connector settings, or add a custom connector with the server URL https://mcp.facebook.com/ads.
3. Sign in and pick the ad account. Approve access with a Facebook login that has permission on the ad account.
4. Start with a read-only question. Ask for last week's spend, results and cost per purchase by campaign, and check the totals against Ads Manager before you trust anything else.

Is there a Google Ads MCP?

Yes. Google released an open-source Google Ads API MCP server on 7 October 2025. It only reads: the GitHub repository lists three tools, for running reports, reading field metadata and listing the accounts you can access, and none of them changes anything in your account.

How to connect Google Ads to Claude

This is a developer job, not a one-click connector. Google's integration guide asks for a Google Cloud project with Google Ads API access, plus OAuth credentials or a service account. You then run the server yourself, either locally or hosted on Google Cloud Run, and point your AI client at it. The repository's setup steps cover Google's Antigravity and the Codex CLI, and explain where the configuration goes for Claude Code, Cursor and VS Code. Google doesn't offer a hosted version for non-developers, so most founders either ask an engineer, use a third-party connector, or use a platform that already pulls Google Ads data.

Meta Ads MCP vs Google Ads MCP at a glance

Why the AI's answers sound right and still mislead you

Ask either connector "what did my ads make last week?" and you'll get a fast, confident answer. The number is real, but it's the platform's number, and five things separate it from what landed in your bank account.

1. Each platform credits itself. Meta counts purchases within its attribution window, by default 7 days after a click or 1 day after a view, as Jon Loomer explains. Google Ads uses data-driven attribution by default. A shopper who clicks a Google ad and saw an Instagram ad the day before can be counted by both, which is how two platforms end up claiming more than the store sold.
2. Some conversions are modelled. Meta says statistical modelling may fill in results when data is missing, and Google's Conversions column includes modelled conversions alongside observed ones. That's sensible for bidding, but it isn't an order count.
3. Purchase value isn't net sales. The value your pixel or tag sends is often the order total, including tax and shipping, and it doesn't fall when an order is refunded. Shopify's net sales is gross sales minus discounts and sales reversals, and excludes taxes, shipping, duties and fees.
4. The dates don't line up. Google's main conversion columns report against the date of the ad click, not the date of the sale, and each ad account has its own time zone and currency. A week in Google Ads isn't quite the same week in Shopify.
5. Nobody has your costs. Neither platform knows your product costs, shipping, payment fees or returns, so a 3.0 ROAS gets presented as a win whatever it leaves you.

Worked example: a 3.0 ROAS that loses money

Take an order Meta reports as $100 of purchase value. $8 of that is shipping charged and $7 is tax, so Shopify's net sales are $85. Product costs are $34, shipping and fulfilment $12 and payment fees $3, which leaves $36 before ads. At a 3.0 ROAS, Meta spent about $33.33 to win it, so the order still makes $2.67. Now suppose 1 in 5 of the orders Meta claims were also claimed by Google, or would have happened anyway. The real ad cost per order becomes about $41.67, and every order loses $5.67. The ROAS in the chat never changed.

You can run your own numbers through the breakeven ROAS calculator, and judge spend against blended numbers like MER and contribution margin, which don't depend on which platform claims the sale.

Letting an AI change your ads: set guardrails first

The Meta connector can act, not just read, and every action needs your approval in the chat. That still means a tired "yes" at 11pm can launch a campaign or move a budget. Before you connect, block campaign creation and cap budgets per ad account in Business Suite, then loosen the rules once you trust the workflow. When something does change, Meta's activity log shows it, and our guide to ad account change history covers how to audit who changed what across Meta, Google, TikTok and Klaviyo.

How to get profit answers instead of ROAS answers

Our Claude Shopify setup guide explains why plugging Shopify and Meta into the same chat still doesn't give you profit: the AI has to stitch together numbers that use different definitions. The fix is the same for Google Ads. Join ad spend with real Shopify orders and every cost, agree one definition of revenue and contribution margin, and only then ask the AI. That's the data step in our AI implementation plan.

It's also what the StoreHero MCP does. One connection gives Claude or ChatGPT your Shopify orders, Meta, Google and TikTok spend, COGS, shipping and fees, already joined and defined, so "which campaigns made money last month?" is answered in contribution margin rather than platform ROAS. Keep Meta's connector alongside it if you want an agent to make changes in Ads Manager.

6 prompts to try once you're connected

• "Show spend, purchases and cost per purchase by campaign for the last 7 days." Check the spend total against Ads Manager or Google Ads first.
• "Which ad sets spent more than $500 this week with no purchases?" Check whether the conversion event is firing before you pause anything.
• "Compare last week's ROAS by campaign with the week before." Check that both weeks use the same attribution setting.
• "What changed in this account in the last 14 days?" Line the changes up against any jump in cost per purchase.
• "Add up the revenue Meta and Google credit to themselves this month." Compare it with Shopify net sales for the same dates. If the ads claim most or all of what the store took, a lot of it is double counting.
• "Which campaigns are above my breakeven ROAS of 2.8?" Work out the breakeven from your own margins first, not the AI's guess.

Key takeaways

• Meta's ads AI connectors (beta since 29 April 2026) let ChatGPT, Claude, Claude Code and Perplexity read and change your ad account through mcp.facebook.com/ads.
• Google's Google Ads MCP is open source and read-only, and needs a Google Cloud project and OAuth to run.
• Both read each platform's own attributed numbers, so Meta and Google together can claim more revenue than your store took.
• Purchase value includes tax and shipping and ignores refunds; Shopify net sales doesn't, and neither platform knows your costs.
• Set guardrails in Meta Business Suite before letting an agent edit campaigns or budgets.
• For profit answers, join ad spend with Shopify orders and costs first, then ask the AI.

Frequently asked questions

Sources

• Meta Business Help Center: Manage ads from an AI agent with Meta ads AI connectors
• PPC Land: Meta opens its ad system to Claude and ChatGPT with new AI connectors
• Google Ads Developer Blog: Open Source Google Ads API MCP Server
• Google for Developers: Google Ads MCP server integration guide
• GitHub: googleads/google-ads-mcp
• Jon Loomer: How the attribution setting works
• Google Ads Help: About attribution models
• Google Ads Help: About conversion modeling
• Shopify: Sales schema (net sales and total sales definitions)