Mockingbird

Explainer

Meta's ads AI connectors: what an AI assistant can do in your ad account

Meta's ads AI connectors are two official ways to let an outside AI tool work in your ad account: a Meta-hosted MCP server for chat assistants such as Claude and ChatGPT, and a command-line tool for developers. Both opened in beta on 29 April 2026. A connected assistant can read your reports and create or edit campaigns, ad sets, ads, audiences and catalogues. Meta's documentation says new campaigns, ad sets and ads are created paused, and turning one on is a separate step.

Last updated

What Meta released, and when

On 29 April 2026 Meta announced Meta ads AI connectors in open beta. The announcement describes them as a Meta-authenticated connection between an ad account and an AI agent, for businesses of all sizes. There are two connectors.

Meta's announcement does not name any AI assistant. Digiday reported on 29 April 2026 that the connectors work with assistants that support MCP, such as ChatGPT and Claude, that they were offered to all eligible advertisers globally, and that access depends on the advertiser's plan with the AI tool. Meta's developer pages give setup steps for Claude Code and ChatGPT.

Two dated additions followed.

  • 16 July 2026. An update on the same announcement page says developers can connect any AI application through their own developer app, and introduces ads MCP server rules. These let anyone with full control of a business portfolio govern what agents may do on an ad account, and Meta gives budget changes and catalogue updates as examples.
  • 6 October 2026. At Advertising Week New York, Meta said the server now lets marketers upload creative, preview ads as they will appear on Facebook and Instagram, and look at performance trends. Meta framed this as being for large advertisers and agencies. The rest of that announcement is covered in our Advertising Week summary.

What a connected assistant can read and change

Meta's tool reference for ad creation and management and for catalogues lists each action the MCP server exposes. Sorting them into reading and changing gives the picture below. The sorting is ours; the actions are Meta's.

What Meta's ads MCP server lets a connected AI assistant read and change
AreaReadChange
Campaigns, ad sets, adsSearch and retrieve them with their performance fields. Meta calls this the main reporting tool.Create a campaign, ad set or ad (each created paused), update fields on an existing one, and switch a paused one to active.
CreativeList creatives, uploaded images and videos, and render a preview for a placement such as Feed, Stories or Reels.Create a creative. The reference describes this tool as single-image link ads only.
InstagramList linked Instagram accounts and their recent posts that are eligible to boost.Boost an existing Instagram post as an ad.
Custom audiencesList audiences, their size and status, and the ad sets that target them.Create or rename an audience, add or remove people on a customer list (as hashed data), and delete an audience.
CataloguesList catalogues, products, product sets, feeds and their diagnostics.Create and update catalogues, products, sets and feeds, delete products, sets and feeds, and connect or disconnect event sources.
AccountList the ad accounts and Pages you can use. The overview also lists signal health, the account's change history and Help Centre search.The overview also lists creating and managing A/B tests and conversion lift studies.

The server can also search the public Meta Ad Library, which Meta's reference describes as being for competitor research and creative ideas.

The Ads CLI covers similar ground. Its overview lists create, read, update and delete for campaigns, ad sets, ads and creatives, plus catalogues, pixels and read-only performance queries.

The controls Meta describes

  • Created paused. The MCP tool reference says the tools that create campaigns, ad sets and ads create them paused, and that the AI client asks for confirmation before activation. It flags the activation tool as the one that starts spending budget. Meta's CLI announcement says the same of the CLI: nothing goes live until you set it to active.
  • Drafts in Ads Manager. On 6 October 2026 Meta said work staged by an agent now lands as a draft in Ads Manager, where people and agents can edit the same campaign and publish from either place. In Meta's words, "Every action is gated behind advertiser-set permissions and explicit approval before anything goes live."
  • Rules. Since 16 July 2026, someone with full control of the business portfolio can set rules on what agents may do on an ad account.
  • Scoped sign-in. The get-started page says the usual route is a Facebook Login for Business screen where you approve permissions, with no token to copy by hand.

How an advertiser connects one

Meta documents three routes. Which one fits depends on how technical you are.

The three routes Meta documents for connecting an AI tool to an ad account
RouteWho it suitsWhat Meta says it needs
MCP server, no app of your ownMost advertisersMeta's announcement says this needs no developer credentials, API setup or coding, and sends readers to a Business Help Centre page for the supported platforms and steps. We could not load that page's text, so we do not repeat its steps here.
MCP server, your own developer appDevelopers and agenciesCreate or reuse a Meta app, add the ads MCP server use case, then add the server address to the AI client. Meta gives a one-line command for Claude Code, and for ChatGPT a developer-mode setup using OAuth and your app ID.
Ads CLIDevelopers, scripts and coding agentsPer the CLI setup page: Python 3.12 or later, a system user with the Admin role in your business settings, the ad account, Page and catalogue assigned to it, and an access token saved in an environment file.

The first route is the one a small advertiser is likely to use. The third hands an admin-level access token to whatever runs the commands, which is a bigger decision than signing in through a chat app.

Sensible precautions for a small account

None of this is Meta's advice. It is what follows, in our reading, from how the tools are described.

  1. 01

    Start with questions, not changes

    Ask for a report on the last 30 days and check the figures against Ads Manager. If the numbers do not match, do not let the assistant act on them.
  2. 02

    Check what it built before you activate

    New campaigns arrive paused. Open them in Ads Manager and read the budget, audience, placements and destination link yourself, as you would in the manual publishing steps.
  3. 03

    State budgets with the currency and the period

    In Meta's own CLI example a daily budget entered as 5000 is annotated as 50.00 in the account's currency. An assistant that gets the unit wrong is off by a factor of one hundred, so confirm the figure in Ads Manager.
  4. 04

    Be explicit about live campaigns

    Tell the assistant which campaigns it must not edit. If you have full control of the business portfolio, set MCP server rules to limit budget and catalogue changes.
  5. 05

    Think before sharing customer lists

    The audience tools can add people to a customer list. That is personal data passing through a second company's AI product. This is general information, not legal advice; check your own privacy obligations first.
  6. 06

    Review the change history

    Meta lists the account's activity log among the things the server can read, and the same history is in Ads Manager. Look at it after each session.

Sources

Questions people ask

What are Meta ads AI connectors?

They are two official tools Meta opened in beta on 29 April 2026 for linking an ad account to an outside AI tool. One is a Meta-hosted MCP server that chat assistants can connect to, and the other is a command-line tool called Ads CLI for developers and scripts.

Can Claude or ChatGPT publish ads in my Meta ad account?

A connected assistant can create campaigns, ad sets and ads, but Meta's documentation says they are created paused and that switching one to active is a separate step the AI client asks you to confirm. On 6 October 2026 Meta also said agent-staged work lands as a draft in Ads Manager and needs explicit approval before it goes live.

Do I need to be a developer to use Meta's ads MCP server?

Meta's announcement says the MCP route needs no developer credentials, API setup or coding, and its developer documentation says owning a Meta app is not required. The Ads CLI is different: it needs Python 3.12 or later and a system user access token.

What can an AI assistant change through Meta's ads MCP server?

According to Meta's tool reference it can create and update campaigns, ad sets and ads, create single-image link creatives, boost Instagram posts, create, edit and delete custom audiences, and create, edit and delete catalogue items and feeds. Since 16 July 2026 a person with full control of the business portfolio can set rules that limit what agents may do.

Try it on your own competitors.

$9 for 7 days. Paste a competitor's Page and see their best ad rebuilt for your brand.

Start 7-day trial · $9

Keep reading

Mockingbird is an independent product from Onwards Analytics and is not affiliated with, endorsed by or sponsored by Meta. Meta, Facebook, Instagram, Meta Ads Manager and the Meta Ad Library are trademarks of Meta Platforms, Inc. Meta changes its tools often; check Meta's own help pages for the current detail before you rely on it.