Meta’s official Ads MCP, in open beta since 29 April 2026, lets Claude and ChatGPT read reporting and edit campaigns, ad sets and ads, while Google’s official Ads MCP still cannot create, change or pause campaigns, budgets or bids. AI now handles much of the execution, but not creative that lifts CTR, scale-or-hold judgment, or accountability for CAC. This guide compares both MCPs, shows when to keep a Claude setup, and gives a break-even test for hiring an agency.
In This Article
A founder told us on a call this week that he had already plugged Claude into his Meta ad account. It had been running for a month. So why, he asked, should he pay an agency when Claude can edit his campaigns for free?
It was a fair question, and the honest answer is more interesting than “you need experts”. The Meta Ads MCP, Meta’s official connector for AI assistants, really does let Claude and ChatGPT change your campaigns. Google’s equivalent still can’t touch a bid. Both facts change what you should pay a human, or an agency like upGrowth Digital, to do.
What follows is what each official ads MCP can and can’t do as of November 2026, checked against Google’s documentation and Meta’s launch coverage. Then come the jobs AI still gets wrong, a checklist for keeping your Claude setup, and a break-even test for when outside help pays for itself.
The Meta Ads MCP lets Claude or ChatGPT read your Meta reporting and create or edit campaigns, ad sets, ads and catalogs through plain-language instructions. Meta launched it on 29 April 2026 as part of Meta Ads AI Connectors, in open beta, alongside a command line tool for developers.

According to PPC Land’s coverage of the launch, the connectors cover 4 jobs: reporting, campaign management, catalog management and signal diagnostics. The same report notes that Meta offered write access from day 1, while Google and Amazon launched their servers with read-only access as the main mode.
Setup needs no developer app. You add Meta’s endpoint as a custom connector in Claude, sign in with the Facebook account that runs your Business Manager, and pick the ad accounts it can reach. That simplicity is exactly why founders are trying it.
Edits made by an agent can trigger Meta’s learning phase, so an AI told to “optimise daily” can keep resetting the campaigns it is meant to improve. Whether an agent flags problems like under-delivery depends on how you configure it. Neither caveat is a reason to avoid the Meta Ads MCP. Both are reasons to put rules around it.
Also Read: Why your Meta ads CPL keeps rising, and how to diagnose it
Google’s official Google Ads MCP server cannot create, modify, pause or enable campaigns, ad groups, ads, budgets or bids. That is stated plainly in Google’s Ads API MCP server documentation, checked on 19 November 2026.
It is not purely read-only any more, though. By default it runs in a “read and partial mutate” mode: it reads account data through search queries, and it can create assets, upload video assets and manage labels. It can apply a label to a campaign. It cannot attach an asset to one.
So the split is uneven. On Meta, Claude can run most of the button-pushing. On Google, Claude can read everything and change almost nothing that moves spend. Anyone telling you AI now runs both platforms end to end is describing Meta and assuming Google.
| Task | Meta Ads MCP | Google Ads MCP |
|---|---|---|
| Read reporting and performance data | Yes | Yes |
| Create or edit campaigns, ad sets (ad groups) and ads | Yes | No |
| Change budgets and bids | Yes, through campaign and ad set edits | No |
| Add catalogs, assets or labels | Catalogs | Assets and labels only |
| Diagnose pixel and signal health | Yes (signal diagnostics) | Read conversion data only |
| Status in November 2026 | Open beta since 29 April 2026 | Read and partial mutate by default |
For execution on a stable Meta account, increasingly yes. For creative, judgment and accountability, no. The Meta Ads MCP automates the tasks an account executive used to do by hand, but it does not decide what to test next or own the result.

Reporting is the clear win. Pulling spend, CTR and cost per purchase by ad set, spotting an ad set that burned budget overnight, and summarising the week now takes a prompt instead of an hour in Ads Manager. Duplicating ad sets, shifting budget between ones you name, and checking pixel health are all reliable on instruction.
The founder on our call named this himself. His click-through rates were low, conversions followed, and Claude was not good at fixing it. That matches what we see. AI can write 20 headline variants, but on Meta the winning ad usually wins on the visual hook and the angle, built from what is converting in your category this month.
Deciding whether to scale a campaign that hit target for 5 days, or hold because the audience is about to saturate, depends on your margins, stock, stage and cash. Claude will give you an answer. It will not know that you can’t fulfil 2x orders next week, unless someone tells it every time.
An assistant does what it is told. Nobody at Claude is accountable when CAC drifts up 30% over a quarter. That is the part of an agency fee that does not show up in a feature comparison, and it is the part founders miss once it is gone.
Also Read: Meta ads agency vs freelancer vs in-house: how to choose
On Meta, targeting has largely moved to the algorithm, so the creative is now the main lever you control. When the system chooses who sees your ad, the ad itself decides whether they stop scrolling.
For a D2C health or wellness brand, that lever is harder to pull. Claims like “relief”, “protection” or anything that sounds medical get reviewed closely, so the angle has to sell without overclaiming. An AI assistant will happily draft copy that gets rejected, and it will not know which compliant angle your competitors are scaling.
This is also where test design matters more than volume. If you want to see how creative testing cadence affects results, our ad creative testing simulator lets you model it with your own numbers.
The best proof we have is not about AI at all. Delicut grew monthly sales from 40K AED to over 2 million AED in Dubai through acquisition, conversion and retention work. Tools changed during that time. The thing that compounded was knowing what to test next.
Keep your Claude and Meta Ads MCP setup if your CTR and ROAS already hit target and someone gives the account 4 to 6 focused hours a week. An agency will not beat a setup that is already working, and a good one will tell you so.

Signs your setup is enough: you ship at least 1 new creative concept a week, you run 1 platform in 1 market, and mistakes at your spend level are cheap. Signs it is costing you: CTR is falling with no explanation, creative changes monthly or less, the founder is also the media buyer, or regulated claims keep getting ads rejected.
Also Read: What a good Meta ads agency includes in scope
An agency pays for itself when its monthly fee is less than the value of the hours you get back plus the extra gross margin from better CTR and conversion. If the right side of that comparison is smaller, keep doing it yourself.

Start with hours. The founder on our call estimated 4 to 6 extra hours a week to run the account properly. That is 17 to 26 hours a month. Value those hours at what a founder hour earns in sales, hiring or fundraising, not at a salary rate.
Then margin. Multiply monthly spend by the ROAS change you expect from better creative and testing, then by your gross margin. If a creative refresh moves ROAS from 2.0 to 2.5 on meaningful spend, the margin line usually dwarfs the hours line. If it doesn’t, the fee is not worth it yet.
Give Claude reporting access first, add write access later, and keep a person approving budget moves and regulated claims. The Meta Ads MCP is powerful precisely because it can change live campaigns, so the controls matter.
Run this way, the Meta Ads MCP is a good junior on your team. It is fast, tireless and literal. It still needs someone to tell it what matters.
The Meta Ads MCP is Meta’s official Model Context Protocol server, launched in open beta on 29 April 2026 as part of Meta Ads AI Connectors. It lets AI assistants such as Claude and ChatGPT read reporting and create or edit campaigns, ad sets, ads and catalogs inside a Meta ad account, using the same login as Ads Manager.
Not yet. Google’s official Google Ads MCP server reads account data and can create assets, upload video assets and manage labels. Google’s own documentation says it cannot create, modify, pause or enable campaigns, ad groups, ads, budgets or bids, so day-to-day Google Ads changes still need a person or another tool.
Only if the gaps cost you more than the fee. The Meta Ads MCP handles reporting and routine edits well. It does not produce creative that lifts click-through rate, decide when to scale or stop based on your margins, or own CAC as a number. If your CTR and ROAS already hit target, keep running it yourself.
It can be, with guardrails. Start with reporting only, log every change the AI makes, and cap how often it edits live ad sets, because frequent edits can push Meta campaigns back into the learning phase. Keep a person approving budget moves and anything in regulated categories such as health or finance.
Creative. Claude can draft copy, but the ads that win on Meta usually win on the visual hook and the angle, built from what is converting in your category right now. Founders we speak to who run the Meta Ads MCP report the same pattern: reporting got faster, but click-through rate stayed low because the creative did not change.
If you already run Claude on your Meta account, you have the data to answer the agency question yourself. Pull CTR by creative for the last 60 days, count how many new concepts you shipped, and run the break-even test above.
If the numbers say your setup works, keep it. If CTR is stuck and creative is the gap, that is the work upGrowth takes on: creative built from what converts in your category, a test plan, and a team accountable for CAC and ROAS. We will tell you on the call which one you are.
In This Article