Meta MCP in Birch AI: benchmark your ads against your industry and turn the gap into a rule

Your CPA went up 18% last week. Your dashboard tells you that much. What it doesn't tell you is whether every other advertiser in your category had the same week.

That difference decides what you do next. If costs rose across your vertical, your creative is probably fine, and pausing it would be the wrong move. If your vertical was flat, something in your account changed, and it's worth finding now rather than at the end of the month.

For a long time there wasn't a practical way to check. The industry benchmarks from Meta aren't exposed through the Graph API and they aren't in Ads Manager reporting. Birch AI reads them through the ads MCP server from Meta, and it can act on what it finds, from pausing an ad to building a new campaign, without leaving the chat. 

Key takeaways

• Birch AI reads the industry benchmarks from Meta and auction ranking data through the ads MCP server—data that isn't available in Ads Manager reporting or through the Graph API.
• Benchmarks tell you whether a change in your costs is specific to your account or market-wide, which changes whether you should intervene at all.
• Auction ranking breaks a performance problem into quality, engagement rate, and conversion rate, so you know whether to work on creative, audience, or everything after the click.
• You can act on what you find in the same conversation—pause or adjust live ad sets, build new campaigns, ad sets and ads, all on your approval.
• Anything you find can become a Birch automation rule that runs on your own thresholds, or a Routine that asks the same question on a schedule, so the same check doesn't need you next week.

In this guide:

• What the ads MCP server adds to Birch AI
• What the native tools from Meta already cover
• Step 1: check your conversion tracking
• Step 2: compare your costs to your vertical
• Step 3: find out which part is underperforming
• Step 4: turn the answer into a rule
• Five rules worth building
• What the connection covers today

What the ads MCP server adds to Birch AI

The ads MCP server is a standard way for AI tools to work directly with the ads platform from Meta. We’ve built it into Birch AI, so there’s nothing separate to buy and no new setup. If you have a Meta ad account connected to Birch, it’s already available—you ask questions the way you already do, and Birch AI now has access to data it couldn’t reach before. 

What it adds: 

• Industry benchmarks. How your CPM, CPC and click-through rate compare to other advertisers in your vertical, and how your campaigns and ad sets compare to aggregated benchmarks from similar advertisers, filtered by spend tier or goal if you ask.
• Auction ranking. Where you land against advertisers like you on quality, engagement rate and conversion rate.
• Account diagnostics. Performance trends by campaign, ad set and ad, your objective and optimization goal, your opportunity score, the change history you see in Ads Manager, alerts when a metric spikes or drops, and an audit of your pixel and event setup.
• Action. Create campaigns, ad sets and ads, and edit live entities — pause, reactivate, rename, change budgets, schedules, bids, optimization goals, and targeting—directly in the conversation.

This page is about using those tools from inside Birch AI. If you want the server itself, what it exposes to any AI client, what it doesn't, and how it differs from the Marketing API, that's in our guide to Meta ads MCP.

What the native tools from Meta already cover

Ads Manager gives you automated rules that can start, pause and adjust budgets on conditions you set. Advantage+ adjusts targeting, placements and budgets during delivery. Opportunity score surfaces setup recommendations. All of that is useful and none of it requires a third-party tool.

What the native rules from Meta can't do is trigger on how your industry is performing. They read your account only. So a rule like Pause ad set if CPA > $40 will pause perfectly healthy ad sets in a week when the whole category got more expensive — because it has no way of knowing the category got more expensive.

Step 1: Check your conversion tracking

Start here. Benchmark comparisons inherit whatever your measurement is doing, so if an event has stopped firing, every comparison you build on top of it is wrong.

Ask Birch AI: "Audit my Meta conversion tracking. Which datasets and pixels are connected, which standard events am I receiving and at what volume, and are any custom conversions broken or no longer mapped to a live event?"

What you're looking for:

• An event that stopped. Purchases fine, InitiateCheckout at zero—usually a site change nobody mentioned.
• A custom conversion pointing at nothing. Common after a funnel rebuild, and it quietly degrades optimization from that day on.
• Optimizing for an event you barely receive. If weekly volume is low, Meta doesn't have enough signal to work with. Fixing this often beats a creative refresh.

Worth running on the first of every month. Tracking doesn't break loudly.

Step 2: Compare your costs to your vertical

Now the question you couldn't answer before: "Compare my CPA, CPM and CTR for the last 14 days against the previous 14 days, and against the industry benchmark for my vertical. Is my gap to benchmark widening, holding or closing?"

The second sentence is what makes this useful. A benchmark on its own tells you very little—plenty of healthy accounts sit above vertical average on CPM because they're deliberately buying a more valuable audience. What's actionable is a gap that moved.

Step 3: Find out which part is underperforming

If the problem is yours, auction ranking tells you where to look.

Ask: "Show me auction ranking versus advertisers like me for my top 5 ad sets by spend—quality ranking, engagement rate ranking and conversion rate ranking. Flag anything below average."

You get three separate reads, and each one points somewhere different:

One follow-up worth asking every time: "For the ad sets with below-average conversion rate ranking—is their CTR actually fine? If so, confirm the problem is post-click rather than creative."

Strong CTR with weak conversion rate ranking is close to conclusive: the ads are working and the funnel isn't. It's also the most common misdiagnosis, because the symptom — rising CPA — looks identical either way.

Acting on it isn’t only about pausing

Diagnosis cuts both ways. When you find a loser, you pause it, but when you find a winner, you can build on it in the same conversation.

With the write tools live, Birch AI can create and edit for you, on your approval:

"Duplicate my best-performing ad set from the last 30 days and adjust its budget."

"Build a new campaign from this brief, targeting the same audience as [campaign], and set it live at a $200 daily budget."

"Raise the budget on the ad set beating the vertical CPC benchmark by 20%."

Nothing goes live until you approve it: Birch AI shows you what it will build or change first. Each Meta MCP tool has its own permission, Needs approval by default, and you can set any of them to Always allow or Blocked. The point is that the whole loop, from spotting the gap to acting on it, stays in one place.

Step 4: Turn the answer into a rule

Everything above is a question you have to remember to ask. If you want to make them stick,  Routines save the question in Birch AI and run it on a schedule, hourly to weekly. The result lands in a chat you can keep questioning, and any change the Routine wants to make waits for your approval unless you've approved that account in advance. A rule goes further and acts on its own, on your thresholds. The benchmark comparison from step 2 works well as a Routine. The pause below needs a rule.

Ask Birch AI to build the rule: "My CTR is running below the vertical benchmark. Draft an automation rule that pauses any ad whose 3-day CTR falls more than 25% below that benchmark, once it has spent at least $100."

Which comes back as rule logic you can review:

Pause ad if CTR (3 days) < vertical benchmark − 25% and Spend (3 days) > $100

Before you switch anything on, ask these three. They're the difference between a rule that helps and one that quietly strangles your account:

1. "How many of my current ads would this rule have paused over the last 30 days?" If the answer is most of them, your threshold is wrong.
2. "Is the spend floor high enough that this won't kill ads before they have data?"
3. "If the benchmark itself moves, does this rule follow it or is the threshold fixed?"

Once a rule is live, you can check on it from outside Birch as well. Birch MCP connects Claude, ChatGPT, Claude Code, or Codex to your rules and their logs, and we've collected what to ask once Birch MCP is connected.

Five rules worth building

What the connection covers today 

So expectations are set correctly: through the ads MCP server, Birch AI reads performance and trends at campaign, ad set and ad level, change history, budget and delivery status, opportunity score, auction insights, industry benchmarks, alerts on unusual movements, and the Ad Library. On the writing side it creates campaigns, ad sets and ads, renames them, pauses and reactivates them, changes budgets (campaign budgets for CBO campaigns), schedules, bid caps, bid strategies, optimization goals, billing events, and audience targeting, and swaps the creative attached to an ad for another one you already have.

On data: benchmarks are aggregated and anonymized. You see how your vertical performs, never a named competitor's account, and your own performance data stays yours.

Wrapping up

If there's one habit worth taking from this: before you rebuild anything, check whether the market moved. A good share of urgent-looking problems aren't yours, and the ones that are get much easier to find once you've ruled the market out.

Open Birch AI and start with the tracking audit, then the benchmark comparison. Ten minutes once the connection is in place.