| Meta Ads | Google Ads | |
|---|---|---|
| Official way to connect AI? | Yes. Meta's Ads MCP Server, open beta since April 2026 | Yes. Google's open-source Google Ads MCP server, since October 2025 |
| Can it change campaigns? | Yes. New campaigns, ad sets and ads are created paused | No. The current release is read-only |
| Rules you agree to | Meta Platform Terms | Google Ads API Terms and Google Ads terms |
| Bans confirmed for using AI? | None. Only anecdotal reports, never confirmed by Meta | None found |
| What does get accounts suspended | Policy and integrity violations | Circumventing systems, unauthorized access, payment issues, policy strikes |
The short answer
No, if you connect through the platform's own route and a human approves every change. Neither Meta nor Google lists "used AI" as a reason to suspend an ad account. Both give you an official MCP server for exactly this.
The risk sits in what the tool does with the access: hundreds of edits overnight, logging in with your password, scraping, or ads that break ad policy. A ban for any of those is a ban for the behaviour, whoever caused it.
Why ad accounts actually get suspended
On Meta
Meta's MCP hub says it plainly: "By accessing or using a Meta MCP server, you and the agent acting on your behalf are bound by the Meta Platform Terms" (Meta MCP hub). Those terms let Meta suspend or remove "your App(s) and account" for violations (Meta Platform Terms, §7.e.ii). Meta's main Terms of Service also ban accessing or collecting data "using automated means (without our prior permission)" (Meta Terms, §3.2.3). Meta's own server is the route Meta built for AI agents, which makes it the safest one.
The same hub warns about prompt injection, "malicious instructions hidden in tool outputs", and that an agent can call anything your granted scopes allow (Meta MCP hub). So the scope you grant matters.
The ban stories are real posts, but they're anecdotal. In early 2026 several advertisers reported Business Managers disabled after bulk overnight changes through third-party MCPs or access tokens (GoMarble, Supermetrics, Porter). One user quoted a restriction notice mentioning "automation that mimics human activity"; the appeal was denied and no Meta staff replied (Meta developer community). Meta hasn't confirmed any of these as AI bans.
On Google
Google's list of suspension reasons is public: egregious violations like circumventing systems and unacceptable business practices, billing problems like suspicious payment activity, unauthorized account access, and repeated policy strikes (Google Ads Help). Using the API, scripts or a third-party tool isn't on it. Google offers automation itself: scripts let you "programmatically manage and query your Google Ads data" (Google Ads scripts).
Two reasons on that list matter here:
- Circumventing systems covers anything that interferes with Google's ad systems, and Google suspends for it without warning (Google Ads policy).
- Unauthorized account access: when Google detects someone else trying to get into your account, it suspends the account to protect against charges (Google Ads policy). A tool that logs in with your password, from its own servers, is the kind of thing that can look like that.
The API has its own rules. Users have to "manually sign in to make changes to their accounts", and tools can't offer "headless programmatic access to make account modifications" without per-account Google sign-in. Scraping is banned (Google Ads API policies). Break those and Google can charge the developer fees, cut its quota or end its API access (same). Google can also end anyone's API access "for any or no reason" (Google Ads API Terms). That's the tool's access, though, not your ad account.
I found no reports of a Google Ads account suspended for connecting an AI tool or MCP, as of October 2026.
Safe vs risky ways to connect AI, side by side
| Safer | Riskier | |
|---|---|---|
| How it connects | The platform's official MCP server or API | A browser bot clicking through Ads Manager, or scraped data |
| How you sign in | OAuth, on Meta's or Google's own login page | Your password, typed into someone else's tool |
| What it can do | Read only, or writes you approve one by one | Changes nobody reviews |
| Pace | A few changes, checked | Hundreds of edits overnight, blind retries on rate limits |
| Scope | One business portfolio or one account | Every portfolio and manager account you own |
| Guardrails | Created paused, Meta server rules, budget caps | None |
How to connect AI safely
1. Use the platform's own server
On Meta that's https://mcp.facebook.com/ads (Meta docs). On Google it's the open-source Google Ads MCP server, which you run yourself, locally or on your own Google Cloud project (Google Ads API docs). Google's is read-only: it "cannot modify bids, pause campaigns, or create new assets" (same).
2. Sign in with OAuth, never a password
Both platforms log you in on their own page and give the tool a scoped token. Nobody else ever holds your password, which is the safest way to stay clear of an "unauthorized access" flag. One 2026 change on Google: developer tokens were sunset on 9 September 2026. API access now comes from the Google Cloud project behind your OAuth credentials (Google Ads API docs).
3. Start read-only
Ask questions first, and only then let the agent change anything. Google's server is read-only already. On Meta, server rules in Business Suite settings can block campaign creation, status changes or targeting edits (Meta, server rules).
4. Connect the smallest scope
On Meta the access covers the whole business portfolio you pick at login. Pick one. On Google, use a login that only reaches the accounts you want the agent to see.
5. Keep a human on every write
Meta helps here: new campaigns, ad sets and ads are created paused, the create tools ask for confirmation, and going live is a separate step (Meta docs). Add a server rule that caps budget increases, for example blocking anything above 20% (Meta, server rules).
6. Check what changed
Meta's server has ads_account_get_activity_logs, the same history as Ads Manager (Meta docs). Ask for it after every session where the agent touched anything.
Where Anry fits
I connected Meta's official Ads MCP to Claude Code on 19 October 2026. The plain URL worked, and it listed 97 tools. Dozens of them can create, change or delete something: campaigns, budgets, custom audiences, catalog products, pixel rules. New campaigns are created paused, which is the right default, but the reach is big.
Anry is the other kind. It only reads your ad data and can't change a campaign. Its own MCP server has 16 tools and none of them can change data, settings, campaigns or ads. It shows where your ad money leaks and what each leak costs in your store's orders, then hands the fix to your agent. You decide.
If all you want is to report on your ads from Claude, you don't need Anry. Meta's and Google's own servers are free.
Frequently asked questions
Can Meta ban my ad account for using Claude?
Using Claude through Meta's own Ads MCP Server is the route Meta built, under its Platform Terms. The reported bans involved third-party tools making bulk changes, and Meta hasn't confirmed them. Meta can still act on whatever the agent does, so keep it read-only or approve each change.
Is there an official Google Ads MCP server?
Yes. Google's Ads API team released an open-source Google Ads MCP server in October 2025, on GitHub under googleads/google-ads-mcp. You run it yourself, locally or on Google Cloud.
Can Claude change my Google Ads campaigns?
Not through Google's official server. It's read-only and can't change bids, pause campaigns or create assets. Some third-party Google Ads MCP servers can write, so check what a tool can do before you connect it.
Do I still need a Google Ads developer token?
No. Google sunset developer tokens on 9 September 2026. Access levels now come from the Google Cloud project you use for OAuth, and the latest Google Ads MCP server doesn't need a token.
Does Google suspend accounts for using automation?
Google's list of suspension reasons doesn't include automation, scripts or the API, and Google offers scripts itself. It does suspend for circumventing systems, unauthorized account access and suspicious payments, whatever the cause.
Are third-party ads MCP servers safe?
Some are fine, but you're trusting them with a token to your account. Prefer OAuth over passwords, read-only over write access, and tools that create things paused and ask before each change.
Sources
- Meta MCP hub, Meta for Developers
- Meta Platform Terms and Meta Terms of Service, Meta
- Ads MCP Server overview, ad creation tools and MCP server rules, Meta for Developers
- Google Ads MCP server, Google Ads API docs, and googleads/google-ads-mcp, GitHub
- Developer token sunset, Google Ads API docs, October 2026
- Google Ads API Terms and Conditions and Google Ads API policies, Google
- Google Ads account suspensions overview, Circumventing systems and Unauthorized account access, Google Ads Help
- Will connecting AI tools to Meta ads get you banned in 2026?, GoMarble (anecdotal reports)
- Anry MCP server docs
See which ad clicks leave your store before they buy.
14-day free trial, no card. Works in Claude, Claude Code and Cursor.