Your agent can analyze your ads. SearchAds Radar lets it act.

One connection gives it three things it cannot reach on its own: your Apple Search Ads account, the subscription revenue that says which spend is actually paying, and a set of write tools that draft changes rather than make them. Nothing reaches Apple until you approve it.

55 tools0 delete tools, by design

Net proceedsafter tax and Apple’s cut, not gross

D0 to D90cohort value, measured not modelled

Re-read at Appleevery write, verified or partial

MCP on freeonboarding tools cost you nothing

Ask what Apple can’t answer.

Find wasted spend, understand what each customer is really worth, and get a plan for what to change next. Your agent connects Apple Search Ads to real revenue, then waits for your approval.

Which spend is producing nothing?

Find campaigns, keywords, and search terms spending money without generating downstream revenue, ranked by wasted spend.

It comes back with14 keywords, $841/mo, zero attributed revenue

Where the $841/mo goesattributed revenue

workout tracker$214$0

fitness app free$187$0

nike training club alternative$126$0

and 11 more under 0.2x$314$0

RecommendsPause 9, negate 5, cut 3 bids by 30%, saving about $612/mo. Waits for your approval.

get_search_terms_analysis get_roas_analysis

Which customers are worth acquiring?

See true ROAS and cohort lifetime value from net proceeds, broken down by campaign, country, and keyword.

It comes back withCompetitor-Terms 2.4x, Brand-US 0.6x, on net proceeds

get_revenue_metrics get_cohort_ltv

What should I change next?

Get a reviewable plan for bids, pauses, and negative keywords. Nothing changes in Apple Search Ads until you approve it.

It comes back withPlan #142: 9 bid cuts, 3 pauses, saves about $612/mo

propose_plan approve_plan

Claude

Audit my last 30 days. Kill anything that doesn't pay.

⏺searchads-radarget_search_terms_analysis›

⏺searchads-radarget_roas_analysis›

I analyzed all 61 keywords against your RevenueCat revenue. 14 of them are spending with zero attributed return: $841/mo in total. The worst offenders:

Keyword	Spend (30d)	Attributed revenue	ROAS
workout tracker	$214	$0	0.00
fitness app free	$187	$0	0.00
nike training club alternative	$126	$0	0.00
…and 11 more below your 0.2 ROAS threshold

My recommendation: pause 9 keywords, add 5 negatives, and cut 3 bids by 30%, that saves about $612/mo without touching anything that converts.

I've drafted this as a plan with propose_plan. Want me to apply it? Every change is re-read at Apple afterwards and reported as verified or partial.

get_search_terms_analysisget_roas_analysis30d · net proceeds

Cursor

fitness-pro - Cursor

PaywallView.swiftStoreKit.swift

struct PaywallView: View {
  @State var plan: Plan
  var body: some View {
    VStack(spacing: 16) {
      Text(plan.title)
        .font(.title2.bold())
      ForEach(plan.perks) { perk
        PerkRow(perk)
      }
      PurchaseButton(plan)
    }
  }
}

AGENT

Cut bids on everything with ROAS under 0.5, show me the plan before you touch anything.

⌄Called searchads-radar · get_roas_analysis

⌄Called searchads-radar · propose_plan

12 keywords qualify. Drafted plan pl_8412: 9 bid cuts of −30%, 3 pauses, estimated save $412/mo.

Nothing changes at Apple until you approve it.

Plan looks right, apply it▍

get_roas_analysispropose_planapproval required

Simulated scene, sample data. See how a change lands →

Your agent sees what happens after the install.

Apple shows the spend and the install. SearchAds Radar connects them to RevenueCat revenue and SARKit attribution, so your agent can measure true ROAS, follow cohort value through day 90, and recommend what to change next.

One customer, from ad signal to real revenueAd signal → Real revenue

01Impression71 of 100 brand searches

02Tap$841/mo never pays back

03Install1 in 3 taps installs

04Subscriber12 customers · $486 paid

05Lifetime valuecohort value through day 90

Your agentGATE 01

SignalYou appear on 71 of 100 searches for your own brand, down from 94.

FindingTwo new bidders are on your brand term and the price floor is rising.

RecommendsRaise the defense bid on brand exact.

ApprovalDrafted, waiting for your approval

Signal$841 a month goes to taps on terms with no revenue behind them.

Finding14 keywords have spent for 30 days and produced no paying customer.

RecommendsPause the 14 and add 5 negatives.

ApprovalDrafted, waiting for your approval

Signal1 in 3 taps becomes an install. This is the last thing Apple can tell you.

FindingInstall rate is healthy, so the loss is happening after the download.

RecommendsKeep bids, follow these installs into revenue.

ApprovalNo change proposed

Signal12 of those installs became paying customers, worth $486 so far.

FindingTwo campaigns produce installs but almost no subscribers.

RecommendsShift budget to the campaigns that convert.

ApprovalDrafted, waiting for your approval

SignalNet proceeds per user at D0, 7, 30, 60 and 90, after tax and Apple’s cut.

FindingTwo countries go flat after day 30 while the rest keep earning.

RecommendsMove spend to the cohorts still growing at day 90.

ApprovalDrafted, waiting for your approval

STANDING BY

GATE 01Impressionread + defend

What goes wrong here

A competitor can quietly buy your brand name and take installs for weeks. No spend report will tell you.

What your agent does

Checks daily how often you actually appear when someone searches your name, against your own history.

What goes wrong here

Cheap taps look like progress. Apple reports them as success whether or not anyone ever pays you.

What your agent does

Scores every term against the revenue it actually produced, then lists the ones producing none.

What goes wrong here

Apple’s reporting ends here. Optimising to installs means optimising to the last number before the money.

What your agent does

Carries each install across the boundary using your own SARKit attribution, no IDFA and no consent prompt.

What goes wrong here

Two campaigns can have identical install costs and completely different revenue behind them.

What your agent does

Joins RevenueCat purchases, renewals and refunds back to the keyword that bought them.

What goes wrong here

A campaign can look cheap per install and never turn into a customer who stays.

What your agent does

Tracks each month’s cohort to day 90 on measured revenue, not a modelled curve.

Campaign

Spend

Installs

CPA

ROAS · net

US Exact

$33.81

$2.60

2.9×

US Broad

$109.79

$2.44

2.1×

Brazil

$15.84

$1.58

0.9×

Everything your agent sees, you can see too.

The panel and the MCP server read the same tool registry, so what your agent reports and what you open cannot disagree. Every change it applies is written to the log with the value before and the value after.