AdPilot MCP
Sign in
Browse documentation
Last reviewed

AdPilot MCP skills for Apple Ads

Skills are ready-to-use workflows that teach your assistant how to complete repeatable Apple Ads jobs with AdPilot MCP tools. A skill explains when the workflow applies, which data to inspect, which operational rules to follow, what a useful answer contains, and when the work is complete.

Use skills when you want a consistent workflow instead of a one-off tool call. You still ask in normal language—the assistant matches your request to an available skill and loads its instructions.

Choose the right skill

Skill Use it when you want to… Can it make a change?
Account orientation Understand what is connected, mapped, current, and ready No
Campaign mapping setup Find and resolve campaigns that are not mapped to an app Yes, internal mapping only
Strategy note capture Save a confirmed decision for future reviews Yes, AdPilot context only
Weekly performance review Compare complete-day Apple Ads performance and explain movement No
Portfolio review Compare apps and decide which one needs attention first No
Search-term waste audit Find irrelevant search-term spend and review exact negatives Only on an explicit apply request
Profitability review Compare Apple Ads spend with RevenueCat proceeds and ROAS targets No
Under-delivery diagnosis Explain why a campaign has low spend, taps, or delivery No
Evaluate previous changes Decide whether an earlier action helped, is still measuring, or should be rolled back Rollback only on an explicit request
Keyword harvesting Turn converting search terms into targeting keywords Only on an explicit apply request
Bid tuning Raise, reduce, or set keyword bids using performance evidence Only on an explicit apply request
Keyword inventory management Inspect, pause, restore, or delete existing targeting keywords Only on an explicit apply request
Negative keyword hygiene Create, review, update, or delete campaign and ad-group negatives Only on an explicit apply request
Campaign launch Build a campaign, ad group, and seed-keyword structure Only on an explicit launch request
Campaign structure management Update, pause, restore, or delete an existing campaign or ad group Only on an explicit apply request
Bounded optimization operator Apply one bounded optimization when no narrower workflow fits Only on an explicit apply request

If two skills seem relevant, start with the narrower one. For example, use Bid tuning for bid changes and Search-term waste audit for exact-negative candidates. Use Bounded optimization operator as a general fallback for a single bounded optimization.

Before using a skill

The assistant can produce a better and safer result when these foundations are ready:

  1. Apple Ads is connected and synchronized. Structure tools need current campaigns, ad groups, keywords, budgets, countries, and status.
  2. Campaigns are mapped to the correct app. App-level reporting and operations depend on that mapping.
  3. The app strategy is confirmed. Protected terms, intent themes, CPA or ROAS targets, and business context guide recommendations.
  4. Spend limits are configured before growth operations. Enabling a campaign or increasing spend requires a workspace monthly spend ceiling.
  5. RevenueCat is mapped before profitability conclusions. Missing or unmapped proceeds are unknown, not zero.

If you are unsure what is ready, begin with Account orientation.

Set up skills

Account orientation

Use this first when the workspace is new, when you have not used AdPilot recently, or when you are unsure which workflow should come next.

  • What it checks: connected apps, campaign mappings, data freshness, confirmed strategy, protected terms, recent decisions, and operation limits.
  • What you receive: a short workspace summary, every important warning, and one recommended setup action.
  • What it does not do: it does not infer missing targets, recommend campaign changes, or execute writes.
  • Example: What is connected to AdPilot, what is ready, and what should I set up next?

Campaign mapping setup

Use this when analytics are incomplete, an app shows unmapped campaigns, or an app-specific tool cannot safely resolve campaign ownership.

  • What it checks: onboarded apps, mapping issues, Apple campaign identifiers, Adam IDs, and app strategy context.
  • What you receive: the exact campaigns that can or cannot be mapped, with the evidence behind each match.
  • When it writes: only when you explicitly ask to apply mappings. The change updates AdPilot's internal campaign-to-app mapping; it does not edit Apple Ads.
  • Example: Show the unmapped campaigns and map the exact Adam ID matches to Calm Budget.

Strategy note capture

Use this when a durable decision should influence later reviews, such as a market priority, an efficiency preference, or a temporary business constraint that remains relevant beyond the current chat.

  • What it checks: the selected app, confirmed strategy, protected terms, and active strategy notes.
  • What you receive: the exact concise note that would be stored, with duplicates removed.
  • When it writes: only when you explicitly ask to save the note. It changes AdPilot context, not Apple Ads.
  • Do not use it for: guesses, unconfirmed targets, short-lived observations, or protected terms that belong in the app strategy.
  • Example: Remember that Calm Budget should prioritize efficient US growth for the next reviews.

Analyze skills

Weekly performance review

Use this for a regular Apple Ads health check or whenever you ask, “What changed?”

  • What it checks: complete-day app and campaign performance, equal-period comparisons, data quality, strategy context, and recent decisions.
  • What you receive: date boundaries, scope, spend and install deltas, the campaigns driving the movement, warnings, and recommended follow-up checks.
  • What it does not do: it separates facts, hypotheses, and recommendations and does not execute writes.
  • Example: Review Calm Budget for the last 7 complete days and compare it with the previous 7 days.

Portfolio review

Use this when the workspace contains several apps and you need to decide where to investigate first.

  • What it checks: workspace-level spend and installs, movement by app, mapping health, data-quality warnings, and lasting decisions.
  • What you receive: a portfolio rollup, top movers, apps with unreliable data, and one recommended drill-down.
  • What it does not do: it does not invent campaign-level revenue or make changes.
  • Example: Compare every app for the last 30 complete days and tell me which app needs attention first.

Search-term waste audit

Use this when you want to find irrelevant queries that spent money without producing installs and assess exact-negative candidates.

  • What it checks: reported search terms, spend, installs, taps, protected terms, intent themes, campaign/ad-group scope, and Apple reporting completeness.
  • What you receive: candidates grouped by campaign with evidence, intent classification, scope, and confidence.
  • When it writes: only when you explicitly ask to apply a candidate. This skill creates exact negatives only, at campaign or ad-group scope.
  • Important limitation: Apple omits search terms below its reporting threshold. Empty or partial search-term data cannot be replaced with invented candidates, and keyword-level waste is not treated as search-term evidence.
  • Example: Audit search-term waste for Calm Budget. Prepare exact negatives, but do not apply them.

Profitability review

Use this when the decision depends on proceeds, payers, ROAS, or payback rather than installs alone.

  • What it checks: Apple Ads spend, RevenueCat proceeds and payers, confirmed financial targets, campaign contributors, attribution quality, and cohort maturity.
  • What you receive: observed ROAS, the target gap, campaign contributors, confidence, and explicit attribution caveats.
  • What it does not do: it never treats missing, stale, or unmapped RevenueCat data as zero and does not execute writes.
  • Example: Review 30-complete-day ROAS for Calm Budget and explain whether the evidence is mature enough to act on.

Under-delivery diagnosis

Use this when a campaign is barely spending, has low taps, stopped serving, or appears constrained.

  • What it checks: mapping, data freshness, campaign status, daily budget, countries, serving reasons, Search Match, ad-group bids, keyword inventory and status, performance, search-term waste, and recent changes.
  • What you receive: likely causes ranked by evidence and a specific next diagnostic action.
  • What it does not do: it does not invent creative, eligibility, or billing failures that are not present in synchronized data. It checks waste before recommending higher bids.
  • Example: Diagnose why Campaign 81 is under-delivering before suggesting any bid increase.

Evaluate previous changes

Use this after an AdPilot action when you want to know whether it helped, harmed performance, is inconclusive, or is still measuring.

  • What it checks: the action record, stored evaluation, evaluation window, before/after metrics, data quality, overlapping changes, and rollback eligibility.
  • What you receive: the original reason, maturity status, measured outcome, confidence, and rollback availability.
  • When it writes: asking Should I roll back? is analysis only. A rollback runs only when the action is reversible, mature evidence supports it, and you explicitly ask to execute it.
  • Example: Did action 54 help? If it is still measuring, tell me when to check again and do not roll it back.

Optimize skills

Optimization skills can change Apple Ads. For a read-only result, use words such as review, analyze, prepare, or do not apply. To execute, clearly say apply, create, update, pause, restore, delete, or roll back, and name the target.

Keyword harvesting

Use this when converting search terms should become targeting keywords in a specific ad group.

  • What it checks: converting search-term evidence, intent, protected terms, destination ad groups, existing targeting keywords, and current negatives.
  • What you receive: non-duplicate candidates with evidence, match type, and destination ad group.
  • When it writes: only on an explicit apply request. Exact match is preferred unless you clearly request broad match.
  • Example: Find converting search terms for Calm Budget and prepare exact-keyword additions for the US Exact ad group.

Bid tuning

Use this when keyword CPA or delivery evidence supports a bid reduction, increase, or exact bid.

  • What it checks: campaign scope, complete keyword inventory, current status and bid, performance, targets, data quality, limits, and recent decisions.
  • What you receive: the current bid, proposed bid, percentage change, evidence, expected risk, and applicable cap.
  • When it writes: only on an explicit apply request. AdPilot validates percentage limits even when you provide an absolute bid.
  • Example: Prepare bid reductions for keywords above my CPA target, capped at 15%. Do not apply them.

Keyword inventory management

Use this for existing targeting keywords, especially when performance reports do not show inactive or zero-metric keywords.

  • What it checks: complete campaign or ad-group inventory, status, bids, performance where available, strategy, and recent decisions.
  • What you receive: a clearly scoped inventory and recommended pause, restore, or deletion actions.
  • When it writes: only on an explicit apply request using exact Apple keyword IDs.
  • Deletion warning: deleting a keyword is permanent. Prefer pausing when evidence is uncertain or the intent may be useful later.
  • Example: List every paused keyword in Campaign 81 and restore only keyword IDs 401 and 402.

Negative keyword hygiene

Use this to inspect existing negatives, fix overblocking, or add carefully scoped new negatives.

  • What it checks: campaigns and ad groups, negative inventory, search-term evidence, protected and relevant intent, recent decisions, match type, and scope.
  • What you receive: exact before/after recommendations for creation, status changes, or deletion.
  • When it writes: only on an explicit apply request. Exact match is the default.
  • Broad-negative warning: broad negatives require an explicit broad-match request and strong evidence that the complete phrase or theme is irrelevant across plausible variants.
  • Removal warning: pausing or deleting a negative can re-open spend. Prefer narrow changes over bulk removal.
  • Example: Review Campaign 81 negatives for overblocking. Prepare fixes, but do not apply them.

Campaign launch

Use this to create a new Apple Search campaign with its first ad group and seed keywords.

  • What it checks: app and Adam ID, Apple organization and currency, payment model, strategy, synchronization readiness, spend ceiling, app discovery, and geo codes.
  • What you receive: an ordered campaign → ad group → keyword plan with exact fields, identifiers, budget, countries, status, and risks.
  • When it writes: only on an explicit launch request. Each structural create executes in order and returns its own audited action.
  • PAYG rule: use a daily budget and omit unsupported lifetime budget fields.
  • Example: Plan a paused US exact-match campaign for Calm Budget with a $20 USD daily budget. Do not launch it yet.

Campaign structure management

Use this for an existing campaign or ad group when you need to change status, name, budget, countries, Search Match, default bid, targeting, or deletion.

  • What it checks: current synchronized structure, performance, serving state, data quality, strategy, recent decisions, geo identifiers, and spend limits.
  • What you receive: verified current state and an exact field-level before/after proposal.
  • When it writes: only on an explicit apply request for the exact campaign or ad-group target.
  • Spend rule: enabling a campaign or increasing exposure requires a configured workspace monthly spend ceiling.
  • Deletion warning: campaign and ad-group deletion must be explicit and may not be reversible.
  • Example: Review Campaign 81 and prepare a daily-budget reduction from $50 to $40. Do not apply it.

Bounded optimization operator

Use this for one bounded optimization when the request does not fit a narrower optimization skill.

  • What it checks: app strategy, data quality, campaign scope, current keyword/ad-group inventory, relevant performance evidence, intent, recent decisions, workspace limits, and stale state.
  • Supported work: one exact campaign negative, keyword pause or restore, keyword bid reduction or increase, keyword creation, or evidence-backed rollback.
  • What you receive: one target, original evidence, exact proposed or returned before/after state, risks, and evaluation window.
  • When it writes: only on an explicit apply request, and it executes one operation at a time.
  • Example: Analyze the most appropriate bounded optimization for Campaign 81. Show me the evidence and do not apply it.

Prepare versus apply

AdPilot does not maintain a separate proposal or approval queue. The wording of your request determines whether the assistant should stop at analysis or call a direct mutation tool.

Request style Expected behavior
Review / analyze / prepare / recommend / should I…? Read data and explain the proposed action; do not mutate
Do not apply / no changes Stay read-only even if a candidate is obvious
Apply / create / update / pause / restore / delete / roll back Execute the named operation after validation and limits pass

After a write, the assistant should report the action_id, exact target, before/after state, provider result, reversibility, and evaluation window. Open Changes for the complete audit record. Your MCP host may also display its own confirmation prompt.

Explore available skills

  1. Open Skills in the AdPilot control panel.
  2. Filter by Set up, Analyze, or Optimize.
  3. Copy an example request or ask for the same outcome in your own words.

Every built-in skill is enabled automatically for every workspace and kept on the latest released version. Skills marked Start here are useful entry points, not separate installations.

Built-in skills still respect the workspace plan. For example, a workflow cannot apply an Apple Ads change when writes are disabled. See Plans, pricing, and subscription limits for the operational entitlements that differ between plans.

The control panel intentionally shows user-facing purpose, outcomes, examples, operational scope, and last-use information. Internal prompt templates, tool sequences, and release versions are not displayed there; a connected MCP client loads the operational guidance automatically when it uses the skill.

AdPilot keeps workspace-specific installation and version records internally for deterministic releases and rollback. Users do not need to manage those records. The Skills page is the source of truth for the available built-in library, examples, operational scope, and last-use information.

How triggering works

You do not need a slash command. A good request names:

  • the app;
  • a complete-day date range when performance is involved;
  • the decision or outcome you want;
  • whether the request is analysis-only or should be applied;
  • exact campaign, ad-group, keyword, or action IDs when you already know them.

For example:

Review search-term waste for Calm Budget over the last 30 complete days. Exclude protected and ambiguous intent, group candidates by campaign, and prepare exact negatives without applying them.

The assistant first lists available skills, matches the trigger, loads the full workflow, and follows its tool order and operational rules.

Skill-use checklist

  • Choose the narrowest workflow that matches the job.
  • Name the app and a complete-day date range.
  • State whether you want analysis, prepared recommendations, or an applied change.
  • Provide exact identifiers when similarly named Apple objects could be confused.
  • Review data-quality warnings, limits, and protected terms before applying.
  • Record the returned action ID after any direct operation.

How AdPilot uses skills

AdPilot exposes available skill metadata to the assistant, then returns the selected workflow guidance for that workspace. Skills organize tool order and completion criteria; server-side authorization and operational limits remain authoritative.

Next steps

Official sources

Need help with AdPilot MCP?

Contact support at support@kovacssoftware.com

AdPilot is an independent service and is not affiliated with, endorsed by, sponsored by, or operated by Apple Inc. Apple and Apple Ads are trademarks of Apple Inc.