AI tools can now build Meta campaigns: what agencies should do about Ads MCP
By Alan Magliocca · Published November 2026
Quick answer: Marketing Dive reported on October 8, 2026 that Meta expanded its Ads MCP server so advertisers can use third-party AI tools to build and manage Meta campaigns, alongside new business assistant features in testing. Before you connect an AI tool to a client's ad account, decide what it may change without a person approving it, start it on reporting, and keep a log of every change it makes.
Key takeaways
- Marketing Dive reported on October 8, 2026 that Meta expanded its Ads MCP server to support more use cases.
- The server lets third-party AI tools build and manage Meta campaigns; Meta first opened it in an open beta in April 2026.
- Meta's developer docs list reporting, creating and editing campaigns, catalogs, signals, tests and activity logs.
- Meta is also testing a business assistant that can change targeting and budgets by chat and send scheduled reports.
- Set approval rules, a change log and a learning-phase rule before any AI tool gets write access to a client account.
What happened
On October 8, 2026, Marketing Dive reported a set of AI tools Meta presented at Advertising Week New York. The report is by Andrew Hutchinson and quotes Meta directly; it names no Meta executives and gives no figures.
Among them, Meta said it expanded its Meta Ads MCP server to support more use cases. The server lets advertisers use third-party AI tools to build and manage campaigns. Marketing Dive's report didn't list which AI tools or which new actions the expansion covers.
The server isn't brand new. PPC Land reported that Meta launched its ads AI connectors, an MCP server and a command-line tool, in open beta on April 29, 2026, with Anthropic's Claude and OpenAI's GPT named as supported assistants. Meta's developer documentation describes a Meta-hosted server that works with any MCP-compatible AI agent and covers:
- reporting and insights;
- creating and editing ads, ad sets and campaigns;
- creating catalogs and fixing feed and item problems;
- signal and dataset health;
- Help Center search;
- A/B tests and conversion lift studies;
- activity logs that mirror the Ads Manager change history.
Marketing Dive also reported that Meta's own business assistant is getting more memory and context, is testing uploads of brand voice and strategy guidelines, can generate creative and change targeting and budgets through chat, and is testing scheduled tasks such as recurring performance reports.
What it means for agency owners
MCP is an open standard for connecting AI assistants to outside tools. With Meta's server, someone on your team can ask an assistant like Claude to pull last week's results or create a new ad set, and the assistant makes the change in the client's account.
That is useful and risky in equal measure. Reporting gets faster. Routine edits can be done in a sentence. But a misread instruction can also change a budget, an audience or a live ad in seconds, in an account your agency is responsible for.
Meta is also automating work agencies used to bill for: making video from static images, writing variations, sending weekly reports. The value you add shifts toward judgment. That means deciding what to test, setting rules the tools must follow, and checking the output.
What to do this week
- Decide who may connect AI tools to client ad accounts, and which tools. Write it down.
- Start every connection on reporting only. Ask the tool for results, and compare its answers with Ads Manager for a week.
- Set an approval rule for changes. Budgets, audiences, new campaigns and anything that spends money need a person to approve before they go live.
- Check how new campaigns are created. Neither Meta's overview page nor the reports we read say whether AI-created campaigns start paused. Test it on an internal account before you rely on it.
- Keep a change log per client. The server's activity logs help, but also record who asked the tool to do what, and when.
- Tell clients. Add a line to your agreement saying AI tools may be used to manage their account under your review.
How AIVX Labs helps
AIVX Labs doesn't connect to Meta's ad accounts, so it won't change a live campaign. The AI Ad Creator writes 3 ad versions for Facebook and Instagram, and the Ad Hook & Angle Generator gives you 20+ angles, which you then review and load yourself. RICO works on the same approval principle described below: it shows the plan and the credit cost for each client job, and runs only when you click Go.
What rules should an AI tool follow in a client's ad account?
A short policy you can adapt:
- Read freely, write with approval. Any tool can pull reports. Any change that affects spend or targeting needs a named person's approval.
- Spend limits. No single change may raise a daily budget by more than an amount you set, for example 20%.
- Respect the learning phase. PPC Land's report noted that frequent edits can restart Meta's learning phase and hurt performance. Limit how often a tool may edit the same ad set.
- No policy-sensitive copy without review. Health, finance, housing and other sensitive categories always get a human read.
- Advantage+ settings stay as approved. If a tool switches creative features on or off, log it. See why an Advantage+ audit still matters.
- One log, one owner. Every client has one change log, and one person checks it weekly.
The idea mirrors the approval rule in Anthropic's background agent, covered in Claude Duet's approval rule: spending money needs a yes.
Which jobs should you give an AI tool first?
Start with jobs where a mistake is cheap:
- Weekly reporting: spend, results and cost per result by campaign, compared with the previous week.
- Diagnostics: checking signal and dataset health, and finding disapproved ads.
- Catalog fixes: spotting items with missing data.
- Drafts: building a new ad set as a draft for a person to review and publish.
Move to live edits only after a few weeks of clean results, and only for the clients whose accounts you know well enough to spot a bad change quickly. If you write the same instructions for several clients, package them once; see Agent Skills as an open standard.
Other software is opening to AI assistants the same way. BrightStage covered how webinars now run from inside ChatGPT and Claude through Zoho's MCP server.
Sources
Written by Alan Magliocca, founder of AIVX Labs and BrightStage AI (Podavinci LLC). Written with AI assistance and checked against the linked sources.
Read our featured article on LinkedIn
Get AIVX Labs verified for $1: RICO, your AI operator, with 7 days of full access.
Get AIVX Labs Verified for $1Follow along:LinkedInYouTubeTikTokXFacebookInstagramSkool community
Frequently asked questions
Which AI assistants can connect to Meta's Ads MCP server?
Meta's developer docs say it works with any MCP-compatible AI agent. PPC Land's April coverage named Anthropic's Claude and OpenAI's GPT.
Do I need a developer to connect it?
PPC Land reported that the MCP path needs no developer credentials or coding, while the companion command-line tool is meant for technical users. Meta's Get started guide lists the exact requirements.
Is it safe to let an AI tool change budgets?
It depends on your rules. Start with reporting, require approval for anything that affects spend, and keep a change log, because a misread instruction takes effect in a live account.
