How I connected Meta's official ads connector to Claude (and what I locked down first)
Written by Benjamin Boman · · 4 min read
Earlier this year Meta launched an official connector that lets AI tools like Claude read (and edit) your ad accounts.
Don’t get me wrong, I like the AI stuff - but this one can also build campaigns, change budgets and switch ads on. So I thought I’d write down how I set it up, in case it gives some useful context before you plug it into a client account.
What it actually is
It’s an MCP server, which is a fancy way of saying a plug that lets Claude talk to Meta’s ads system directly. You ask a question in plain English (“which ad sets spent the most last week, and what did a lead cost?”) and Claude pulls the numbers for you. No exports, no clicking through Ads Manager.
Sounds great, right?
But Meta’s AI account security is extremely volatile right now. Even the OGs are experiencing it. Most of the ban stories earlier this year involved third-party connectors and shared developer apps, not Meta’s own one, and Meta’s help centre says an account is unlikely to be banned just for using it. (That’s not a promise though. Your ads still get reviewed like any other ad, and the usual reasons ad accounts get banned still apply.)
It’s still in open beta, so it might not have reached your account yet.
Setting it up
- In Claude, go to Customize > Connectors, click the +, then “Add custom connector”. Call it something like “Meta Ads” and paste in
https://mcp.facebook.com/ads - Click Connect and log in with Facebook. You’ll be asked which business portfolios it can see. This is the important bit (more below).
- Start a new chat, click + > Connectors and make sure it’s switched on.
It works on Claude’s free plan too, but free accounts only get one custom connector at a time. On Team or Enterprise plans, the organisation owner has to add it first.
Lock it down before your first chat
- Check which portfolios you ticked. The connector sees everything your Facebook login can see. That includes client accounts shared with your portfolio, and old clients you forgot to offboard. I run client accounts as a white-label contractor, so a mistake here lands on someone else’s pixel and history, not mine.
- Set the Ads MCP Server rules. In Business Settings > Integrations > Ads MCP Server you can choose, per ad account, which actions are allowed, blocked or need approval. The default has been seen as everything allowed. I start with budget changes blocked.
- Set Claude’s own permissions. In the connector settings, put write tools on “ask” and delete tools on “block”. That way nothing changes without you clicking yes.
- Keep everything paused. New campaigns and ads go in paused by default. I review them in Ads Manager and switch them on myself.
What I use it for
- Checking for anything rejected, broken or not delivering
- Spend and cost per lead by ad set for the week
- This month vs last month, without exporting anything
- Drafting a new campaign as paused, so I just review and switch it on
If you’re new to reading those numbers, here’s how I analyse Facebook ads results, even with low conversions.
What I don’t let it do (yet)
- Change budgets or bids
- Switch anything on
- Delete anything
And one more. If a request fails, stop and look at the error rather than asking it to try again and again. I can’t prove it, but my guess is lots of rapid retries is exactly the kind of pattern Meta’s systems flag. And it’s easy to do by accident when you’re chatting.
Is it worth it?
For reporting and quick checks, I think so. I’d start there for a couple of weeks.
For building campaigns, it’s probably fine as long as everything lands paused and you check it before it spends a cent. But it’s early days, who knows.
If you’ve been using it for builds already (or decided not to), I’d love to hear how it’s going.
Sources: Meta’s Ads MCP server docs and Jon Loomer’s setup guide.