Last week, Meta announced AI Connectors, which allow advertisers to manage campaigns directly through their existing AI tools, using Meta’s ads MCP server. At the moment, AI Connectors work for Claude, ChatGPT, and Perplexity.
Today, let’s focus specifically on using an AI Connector for Claude. In this post, I’ll provide details on how to set up the connector in Claude to get started before discussing some specific use cases.
A couple of notes before we proceed:
First, since Meta announced AI Connectors as an open beta, not all advertisers currently have access. You may be able to complete the steps below, but getting data from the connection may not yet be available to you.
Second, there are very real risks involved, including the possibility that connecting AI tools to your account could draw the wrong kind of Meta attention. You should use caution, and I’ll close this post by reiterating those potential risks.
Let’s get to it…
Connect to Claude
You don’t need a premium Claude plan to set this up. Custom connectors are available on every plan, including the free tier. Free users, though, are limited to one custom connector at a time. If you’re on a Team or Enterprise plan, your organization Owner needs to add the connector first before you can connect to it.
While logged into Claude, click “Customize” in the left menu.
Click “Connectors.”
Click the “+” to add a new connector and select “Add custom connector.”
Name your custom connector and provide Meta’s remote MCP server URL (https://mcp.facebook.com/ads). Then click “Add.”
Your custom connector will be added to the list.
But you won’t be connected yet.
Click “Connect.”
You’ll then be taken through Facebook Login and a series of prompts to select the business portfolios that you want connected to Claude.
When you start a new chat, click “+” > Connectors and make sure that the connector you just set up is toggled on.
Tools Available
The capabilities of this connector are driven by the tools available to it. Once you’ve completed the connection, there’s a list of the tools Meta allows Claude to use. I’m currently seeing 29 of these tools.
I won’t list them all, but here are a few examples:
ads_get_ad_entities — Pulls performance data at the account, campaign, ad set, or ad level for any time period, with filters and breakdowns.
ads_insights_advertiser_context — Provides a high-level overview of your business and marketing funnel to give Claude context for the rest of the conversation.
ads_insights_performance_trend — Analyzes the direction of your key metrics over time.
ads_insights_anomaly_signal — Surfaces unusual patterns or deviations in performance that may warrant investigation.
ads_insights_industry_benchmark — Compares your ad set performance against aggregated benchmarks from similar advertisers.
ads_insights_auction_ranking_benchmarks — Highlights which of your ads are winning the auction and which are getting under-delivered.
ads_get_errors — Surfaces delivery-blocking errors at the campaign, ad set, or ad level.
Use Cases
All interesting, but what can you actually do with this? What makes it worth using? I’d understand if this remains unclear to you.
Whenever you’re unsure of the best ways to use Claude for something like this, there’s no better way to uncover what’s possible than by simply asking it. And so I asked Claude for a handful of ways this AI Connector could be used, including a specific example for each.
Here’s what Claude returned (with some light edits)…
1. Account Exploration and Reporting
This is the most obvious starting point. You can ask Claude to pull performance data for any time period, at any level (account, campaign, ad set, or individual ad). Sort by spend, ROAS, CTR, frequency, or any other metric. Filter by status, objective, or thresholds you set. Apply breakdowns by age, gender, placement, country, device, and more.
Where this differs from Ads Manager is the back-and-forth. Ads Manager is fast if you already know which columns and filters you want. The connector replaces the moments where you’d normally export to a spreadsheet and build a pivot table to answer a slightly more nuanced question. Ask questions using natural language to get quick answers.
Specific example: Pull your top 10 ad sets by ROAS over the last 30 days, but only ones with at least 50 conversions, broken out by placement.
2. Diagnostic Work
Use this connector for troubleshooting. Claude can surface anomalies like sudden CPM spikes, frequency creep, and conversion drops that you might not normally catch. It can analyze auction performance to identify which of your ads are winning or losing in delivery against competing advertisers. It can also pull benchmarks so you can determine whether your performance is in line with similar advertisers in your category.
Specific example: Ask Claude what’s unusual about your account this week, then dig into any anomalies it finds.
3. Pre-Flight Error Checking
Meta surfaces delivery-blocking errors across campaigns, ad sets, and ads — the kind of red-flag issues that prevent things from running. The connector lets you ask for a sweep across your entire account in one prompt instead of clicking into each entity to check. This only catches configuration errors, not policy rejections or account-level restrictions.
Specific example: Ask Claude to surface any delivery-blocking errors across all active campaigns and ad sets.
4. Signal and Account Health
Three tools cover dataset and pixel health: configuration metadata, signal quality scores, and event volume statistics. This is the audit work that’s normally tedious, like checking whether your pixel is firing properly, whether match quality is good, or whether events are coming through at the volume you’d expect. This pairs well with Conversions API setup as a way to verify it’s working.
Specific example: Ask Claude to review your pixel health and event match quality, then summarize anything that looks off.
5. Campaign and Ad Creation
Claude can create campaigns, ad sets, and ads, all of which are created in PAUSED state by default. Nothing goes live without you flipping the switch in Ads Manager. It can also update existing entities (budgets, names, targeting) and activate or pause things on request.
Specific example: Ask Claude to duplicate your best-performing ad set with three new creative variations, all paused for review.
6. Catalog Creation and Management
If you run ads for e-commerce, the connector can create new product catalogs, browse and search products and product sets, and pull diagnostic information on data feeds and item visibility issues. It can also surface details on individual products, feed configuration, and ingestion rules.
Specific example: Ask Claude to identify any products in your catalog with feed errors or visibility issues, then summarize what’s causing them.
The Risks
The risks are real, and they start with whether you should set up this AI Connector in the first place.
It was only about two months ago when advertisers started reporting unexplained account shutdowns in response to unapproved integrations between Meta and AI tools. And now Meta’s inviting us to do it?
Granted, this shouldn’t be a trap. The account shutdowns were likely in error (though I haven’t heard what’s become of those accounts), and Meta has now announced these official integrations. We should be safe now, but I’m just as nervous as anyone.
There are other potential risks related to automations. New campaigns, ad sets, and ads created through the connector are put in a paused state by default, and you’d need to activate them from Ads Manager. Any tool that has full read and write access should make you nervous. How much damage can be done from a bug?
Another potential risk is the connector inherits all your Facebook user’s account access. Your business portfolio likely has access to ad accounts given to you by clients. That could generate additional business risks.
And finally, there’s a risk related to expectations. This is amazingly powerful and impressive technology. There are countless possibilities of what you could do with it. Using it for reporting, troubleshooting, and generally uncovering information is low risk. It gets riskier once you think that automating your processes, ad creation, and edits with little oversight means results will definitely be better.
The potential is huge. But, a whole lot of advertisers are learning some valuable lessons about AI. And those lessons can be expensive.
Some of these possible risks are related to unknowns because this is new technology. Some of this will likely be clarified shortly. Use caution. Be smart. And if it means avoiding the implementation of AI Connectors until you feel comfortable, wait.
FOMO is better than losing your shorts.
Your Turn
Have you set up a connection to your business portfolio using AI Connectors yet? How are you using it?
Let me know in the comments below!
