Meta just opened Ads Manager to Claude — here's the setup that actually works
Meta just shipped an official MCP server that lets Claude, ChatGPT, and Gemini drive your ad accounts through natural language. Here's what it does, how to wire it up in five minutes, and what's quietly waiting to bite you if you do this without knowing the platform.
Meta quietly shipped something big two weeks ago. On 29 April 2026, they opened up the Ads AI Connectors program — an official Model Context Protocol (MCP) server that lets AI assistants like Claude, ChatGPT, and Gemini talk directly to your Meta ad accounts. No developer app. No Marketing API approval queue. No third-party token broker sitting in the middle. Just OAuth, and you're in.
I've been waiting for this for a while, and I think a lot of you have too.
The shift this actually represents
The marketing operations stack has been bloated for years. Five tools to plan a campaign, three to launch it, two to report on it, and one more for everyone to argue about what the numbers mean. The cost of all that friction shows up as either a fat agency retainer or a busy in-house team that spends most of its week copying numbers between spreadsheets.
MCP changes the shape of the answer to that problem. Not because the AI replaces the marketer's judgement — it absolutely doesn't, and I'll come back to that — but because the gap between what you want to do and the thing that actually happens in the ad account drops to seconds. One operator with deep platform knowledge, sitting in front of one AI assistant, can now run more accounts more carefully than a team of five could three years ago.
I think the market is going to move in that direction faster than most agencies are prepared for. Instead of paying a roster of mid-level account managers to navigate Ads Manager all day, you'll increasingly hire one senior operator — someone who actually understands media buying, attribution, and creative testing — and they will run multiple accounts through an AI layer. Fewer hands on the keyboard, more brain per dollar of media spend. The agency model that survives this shift will be the one that figures out how to package senior judgement and AI tooling together, not the one that keeps selling seat-hours.
What Meta actually shipped
The MCP server lives at https://mcp.facebook.com/ads. It exposes around 30 tools split roughly into three buckets.
Read tools do exactly what you'd expect — list ad accounts, campaigns, ad sets, and ads; pull insights with the full breakdown matrix (country, device, age, placement, publisher platform, hour of day); fetch creatives; query opportunity scores and benchmarks; surface auction-ranking diagnostics. You can ask for the last seven days of performance broken down by country and get a clean table back. You can ask "which of my ad sets have rising CPM and frequency over 4" and get a fatigue list ready to act on.
Write tools are the part that actually changes how you work. The MCP can create campaigns, ad sets and ads; update budgets and statuses; activate or pause entities; and manage catalogs and product sets. You'll be able to say "pause every campaign with ROAS under 2 over the last 14 days" — and have it executed. You'll be able to say "find the three best-performing creatives from last quarter and clone them into the Italy account with the budget split 60/30/10" — and have it done.
Diagnostic and signal tools round out the set: pixel and dataset quality checks, anomaly detection, performance-trend analysis. These are the things that today take three browser tabs and a coffee. They become one prompt.
The thing that matters about all this is that it's officially Meta, on Meta's infrastructure. No rate-limit risk. No "your account got flagged because a homemade script touched it" risk. No long-lived access token to babysit. The platform is sanctioning automation directly, which is the part of this announcement most people are sleeping on.