For years, anyone who wanted to truly automate Meta Ads had two options, and both were inconvenient. They could pay for a third-party connector, with the constant concern that the account might one day get flagged, or build an integration with Meta’s APIs, along with everything that entails, from a developer app and approvals to waiting periods. On April 29, 2026, Meta changed the rules: it launched the Meta Ads AI Connectors in open beta, an official way to connect assistants such as Claude and ChatGPT directly to an ad account and have them work through natural language.
The launch has gone somewhat unnoticed in Italy, where it’s still rarely discussed and almost always in the context of tutorials. Yet it’s one of the most important changes of the year for anyone managing campaigns because it shifts where the work happens. No longer inside Ads Manager, but inside a conversation.
What Are They, in Simple Terms?
The connectors are two tools that communicate with the same underlying engine. The first is an MCP server, meaning a service hosted by Meta that uses the Model Context Protocol, the open standard that allows AI assistants to connect to external systems. You add it to Claude or ChatGPT as a connector, sign in with your Meta account, and the assistant can access the ad account. The second is a CLI, a command-line tool designed for people who work in the terminal, for example with Claude Code, and for those who want to run processes automatically without anyone sitting in front of a screen.
There are around thirty functions covering four areas: performance reporting, campaign management, catalog management, and signal diagnostics. In practical terms, the assistant can read spend, impressions, clicks, conversions, frequency, and ROAS at every level of the account. It can create and modify campaigns, work with product catalogs, and check the status of pixels and signals. Both read and write access are supported.
Meta wasn’t first. Google had already released an open-source MCP server for Google Ads in October 2025, and Amazon Ads had opened its own in closed beta in November of the same year. But Meta was the more closed platform, the one that had forced agencies and freelancers to rely on unofficial tools for years. The fact that it has now opened up is a signal that the market has already decided where things are heading.
What They Do Well
The biggest thing connectors do well is shorten the time between asking a question and getting an answer. “Which are the ten ads with the best ROAS over the past fourteen days?” “Which ad sets have exceeded a frequency threshold that’s already too high for this client?” “Which creatives are losing CTR week after week?” These are questions a media buyer asks every day, and they normally require filters, custom columns, exports, and a spreadsheet. Now they require a sentence.
The second use, for anyone managing multiple accounts, is reporting. A process that reads the previous day’s data every morning, compares it with the average of the past few weeks, and flags only the anomalies is exactly the kind of thing the CLI is designed for. Anyone who works like I do across multiple clients in parallel knows how much time is spent not analyzing data, but collecting the data that needs to be analyzed.
The third, and perhaps most underestimated, use is signal diagnostics. I’ve written extensively about how the problem with many accounts isn’t Meta Ads itself, but what Meta can no longer see. Being able to ask an assistant to check event quality, parameter matching, and the status of the Conversions API in a few seconds turns a check that often gets postponed into a daily habit.
What They Can’t Do, and It’s Not a Technical Limitation
This is where some honesty is needed. A connector changes how quickly you can ask your account a question. It doesn’t change the quality of the question. If you ask the assistant which campaigns have the lowest CPA, it will tell you accurately. It won’t tell you that the CPA is low because the campaign is generating poor-quality leads that the sales team rejects on the phone, because that information doesn’t live in the ad account. It lives in the CRM, in notes from the office, and in the head of the person making the calls.
It’s the same mistake I’ve seen in reporting for years: confusing the thermometer with the fever. The assistant reads the thermometer extremely well, and very quickly. But diagnosis requires knowing things that aren’t in the account: margins, client seasonality, actual business objectives, commercial constraints, and the difference between a lead who buys and one who is simply asking for the price.
Then there’s the issue of making changes. An assistant that can create and modify campaigns is a powerful tool, and like all powerful tools, it should be used according to one simple rule: the machine proposes the changes, a person approves them. Not because of distrust toward AI, but because an error in a daily budget or audience exclusion costs real money, and in a client’s account, it also costs trust.
The Bigger Picture: Agents on Both Sides
There’s a reason I find this launch more interesting than it might initially seem. In the same year that Meta is opening its advertising system to advertisers’ agents, it’s also bringing a personal agent such as Muse to users. Put another way: on one side, an agent buys the advertising; on the other, an agent reads it and decides what reaches the person. Marketing is becoming, at least in part, a conversation between machines, and the companies that figure out how to communicate with both sides first will have an advantage that could last for years.
This doesn’t make the media buyer irrelevant. It makes the role different. Less time spent navigating menus, more time deciding what to ask, what to measure, and when to reject a recommendation that looks great in the numbers but terrible for the business.
Where to Start
If you manage a single account, start with read access. Connect the connector, ask the questions you normally ask every Monday morning, and compare the answers with what you see in Ads Manager. You’ll quickly learn where the assistant is reliable and where it needs more context. If you manage multiple accounts, the next step is automated anomaly reporting, not automated campaign creation. And in any case, before giving the agent permission to modify anything, ask yourself whether you could explain to a client why that change was made.
This time, the technology arrived before the processes. The question worth asking isn’t whether to use Meta Ads AI Connectors, because in a year almost everyone will be using them. It’s another question: when anyone can manage an account with a single sentence, what will still make your work different from everyone else’s?
