Meta Ads MCP server

The active Meta ad account: name, id, currency, and whether it can still spend. A disabled account reports healthy numbers right up until it stops delivering, so this is the first call before trusting any performance read.

Every Meta ad account this workspace can act on, with its currency, and which one is active.

Change which Meta ad account subsequent calls operate on. The choice persists. Changes no ad-account state.

Get high-level KPI metrics for the ad account: spend, impressions, reach, clicks, CTR, CPM, CPC, frequency, leads, CPL. For Sales/e-commerce accounts the payload also includes result_type:'purchases' with purchases, cost_per_purchase, roas, and revenue — for those, ROAS is the primary metric, not leads/CPL. Use this first to understand overall account health.

List campaigns with their name, status, objective, budget and budget_type (CBO = budget on the campaign, ABO = budget on each ad set). Returns campaign IDs needed for deeper analysis. truncated: true means the account has more campaigns than were returned.

Get per-campaign performance: spend, impressions, CTR, CPM, CPC, reach, leads, CPL. Sales/e-commerce campaigns additionally return result_type:'purchases' with purchases, cost_per_purchase, roas, and revenue — compare those campaigns on ROAS/cost per purchase, never on leads (a Sales campaign showing 0 leads is NOT broken if it's driving purchases). Defaults to ACTIVE campaigns only. Use this to compare campaign performance and find winners/losers.

Get ad-level performance breakdown for a SPECIFIC campaign. Returns each ad's spend, CTR, CPM, CPC, frequency, leads, CPL, and current status. For ads in a Sales/e-commerce campaign, each ad also carries result_type:'purchases' with purchases, cost_per_purchase, roas, and revenue — rank those ads by ROAS/cost per purchase, not leads. Use this when the user asks about a specific campaign's ads, creative performance, or wants to drill into one campaign.

Get all ad sets inside a campaign, with their status, daily budget, and targeting info.

Get performance metrics broken down by ad set within a campaign. Ad sets in a Sales/e-commerce campaign also include result_type:'purchases' with purchases, cost_per_purchase, roas, and revenue — judge those on ROAS, not leads/CPL.

Get individual ads inside an ad set with full creative copy: name, status, primary text, headline, description, CTA, link URL, thumbnail, and creative type. For dynamic creatives, returns all body/headline/description variations.

Find the best-performing ads in the account by spend and conversions. Good for identifying winning creatives. Ads from Sales/e-commerce campaigns also carry result_type:'purchases' with purchases, cost_per_purchase, roas, and revenue — for those, the 'best' ad is the one with the strongest ROAS, not the most leads. 'today' is a PARTIAL day and 'yesterday'/'today' are tiny samples — report them as snapshots, never declare winners or losers from them; prefer last_30d for decisions.

Statistically proven winners, grouped by creative dimension: which primary text (copy), headline, landing page, or video hook actually performs best. Unlike get_top_performing_ads (individual ads ranked by spend), this aggregates every ad sharing the same copy/headline/page/hook and ranks groups by real cost per result — with a 95% z-test confidence flag vs the account average, a minimum-results trust floor so 1-lead flukes can never rank as winners, and a settled-window CPL trend (improving/worsening). Purchase-optimized accounts rank on ROAS/cost-per-purchase automatically. USE THIS for 'what copy/headline/hook is winning', 'what should we make more of', or creative strategy questions.

Get the account's daily briefing: AI narrative + headline, mood, week-over-week deltas, current winners and bleeders, budget recommendations, and alert/fatigue summary. Numbers come from a SETTLED 7-day-vs-7-day window (ends yesterday — today's partial day is excluded so attribution lag can't fake a decline), and results count each ad set's real optimization event (custom conversions included), never name-guessed leads. Purchase-optimized accounts read as purchases/cost-per-purchase. Served from a 30-minute cache when fresh; pass force:true to regenerate. USE THIS for 'morning briefing', 'how are we doing', or any daily status question.

Calculate AutoAdy's ROI for this account: net money saved from pausing bleeders (reconciled against the account's own spend), time saved, performance improvement, and an ROI multiplier of net savings over the plan price. The multiplier is 0 when nothing was saved; time saved is priced separately as time_value, in the account's currency, at labor_rate_per_hour (default 50 USD an hour, converted). roi_basis states how both were computed.

Run a read-only Meta Marketing API (Graph) request and return the JSON. USE THIS when the user asks for a Meta field or edge that the named tools do not expose. The path is a node optionally followed by one edge, for example "act_123456/campaigns" or "23848.../insights". Readable edges: adcreatives, ads, adsets, campaigns, customaudiences, insights, leadgen_forms. Costs 4 credits per run on the Solo, Growth and Agency credit plans; existing Pro and Agency subscriptions get 100 and 500 runs a day free instead, per workspace, resetting at midnight UTC. Not available on the free plan. Prefer a named tool when one exists. This tool cannot change anything.

Run a comprehensive audit of the entire ad account. This fetches account-level insights, all active campaigns with their metrics, and ad-level data for the top 5 campaigns by spend. USE THIS FIRST when the user asks for an audit, analysis, or account review. The returned payload contains everything needed for structured recommendations. Sales/e-commerce campaigns include result_type:'purchases' with purchases, cost_per_purchase, roas, and revenue — judge those on ROAS and cost per purchase, NOT leads/CPL.

Diagnose WHY account performance broke (CPL spiked, leads dried up, results collapsed). Runs 4-5 independent blind investigators over 35 days of daily data — creative fatigue, delivery/budget shifts (with the account's confirmed budget-change log as facts), traffic quality, offer/seasonality, and conversion tracking — then an adversarial pass tries to kill each theory. Returns a ranked cause-vs-symptom diagnosis with evidence for/against and the single most important action this week. USE THIS when the user asks why performance dropped or what broke; it prevents the default wrong guess (killing a winning creative for 'fatigue'). A full run takes about 2 minutes. Over MCP the call answers within about 20 seconds, and if the run is not finished it returns status "running" with retry_after_seconds — call it again with the same break_description to get the stored diagnosis.

Compute the account health score (0-100) with 5 components: creative diversity, fatigue risk, budget efficiency, testing velocity, waste level. Returns overall score plus component breakdowns.

Get a strategic synthesis of all pre-computed intelligence: alerts, forecasts, recommendations, benchmarks, budget allocation, and priorities. Use this when the user asks 'what should I do?', 'give me a strategy', 'what's the plan?', or any high-level strategic question. This does NOT call Meta API — it reads from our intelligence engine's cache.

Get creative fatigue predictions from the survival analysis engine. Shows which ads are fatiguing, how many days remain, and which need replacement. Reads from the intelligence cache.

List the open findings from AutoAdy's continuous account scan — the same actionable to-do queue shown on the dashboard (bleeders to kill, winners to scale, anomalies, tracking issues). Each finding has a severity, plain-English rationale, a proposed action, and an expected impact measured over the finding's own data window (NOT a monthly projection). Findings auto-resolve when the underlying entity recovers, so this list is live, not a stale report. Use resolve_finding to mark one handled.

Mark one scan finding as resolved/handled so it leaves the findings queue (on the dashboard and in get_findings). This changes AutoAdy app state only — it never touches the Meta account. Use after the user has acted on a finding (or explicitly dismisses it). Pass the finding id from get_findings.

The account's decision ledger (proof of work): every action AutoAdy or the agent executed — pauses, budget changes, launches — with when, why (source), the before-snapshot, and the MEASURED impact where computed (CPL/leads/spend before vs after, and a verdict). USE THIS for 'what did AutoAdy do', 'did that budget change work', accountability questions, or before repeating an action that was recently tried. Every write tool logs here automatically.

Analyze creative performance by visual style, messaging angle, or hook tactic. Returns top and bottom performers with scores, strengths, weaknesses, and improvement suggestions. Use when the user asks 'what creative styles work best?', 'which hooks convert?', or anything about creative tag performance.

Creative analytics for the currently selected, authorized ad account: win rate and cost-per-lead by messaging angle, hook, visual style and format; the angle-by-style combo matrix; 28-day cost trends per angle; and measured review suggestions. Use when asked what creatives/angles/hooks work or what to test next. Report coverage and missing evidence; do not infer that creative changes caused a cost change. Select another authorized account before reviewing its analytics.

Review framework evidence within the currently selected, authorized ad account. This scoped tool does not read other accounts or establish cross-account patterns. It preserves the multiple-account evidence floor and reports cross-account evidence as unavailable when it cannot be established. Use get_creative_analytics for this account's measured angle, hook and style patterns; never describe scoped results as a winning portfolio playbook.

Deep analysis of a single ad: returns the actual creative image/video URL, full ad copy (primary text, headline, description, CTA, link), adset targeting (age, gender, locations, interests, behaviors, custom audiences, placements), adset budget + optimization goal, and campaign objective + budget type. Use after get_top_performing_ads to understand WHY a winner works before multiplying it. If it's a video ad, returns the video source URL.

AutoAdy's media-buying playbooks: foundation, audit, diagnose, creative strategy, launch, targeting, scale.

The full methodology behind one playbook, so the assistant works from an opinionated process instead of improvising.

Plan HOW to scale a winner — vertical (raise budget), horizontal (duplicate to a new audience), bid-cap (Tichenor), or target-ROAS (Faris) — and return the exact execution parameters (budget step, bid amount, duplication method). Pass axis='auto' to let it pick vertical vs horizontal from frequency/CPA-inflation. This is advisory: it returns the plan; execute vertical via scale_budget_relative and horizontal via duplicate_adset. Use with the Scale skill.

Get a multi-week tactical playbook (dated week-by-week checklist) for a campaign window. Call with no slug to list available templates: bfcm, product-launch, lead-gen-ramp, recovery, scale-ladder. Use with the Playbooks skill when the user wants a BFCM/launch/recovery/scaling plan.

Replay a draft rule over the account's last 30 days of real data BEFORE creating it: how many times it would have fired, which ads it would have hit, and (for pause rules) the spend it would have avoided. ALWAYS run this before create_rule and show the user the result — 'this rule would have fired 3× last month on: …' — so they confirm on evidence, not on a description. Read-only, changes nothing.

List the Instant Lead Forms that already exist on the account's Facebook Page, with their ids, names and status. Use this to REUSE an existing form (pass its id as lead_gen_form_id to create_lead_ad / create_lead_ads) instead of creating a duplicate — matching a client's proven form matters because leads from a new form land in a separate Meta form export.

List the account's custom + lookalike audiences: id, name, subtype, approximate size, delivery status (read-only). USE THIS before creating audiences (avoid duplicates) and to get the ids for create_adset targeting (custom_audience_ids / excluded_custom_audience_ids) or the seed for create_lookalike_audience.

List the Meta Pixels (datasets) on the ad account: id, name, and when each last fired (read-only). USE THIS to find the pixel_id for create_adset with OFFSITE_CONVERSIONS — e.g. building a website-conversion Leads campaign — so the user never has to dig it out of Events Manager. A pixel that has NEVER fired, or not fired recently, is a warning: Meta optimizes poorly against an event it hasn't seen — verify the installation (audit_tracking) before launching on it.

Adjust the daily budget of a CBO campaign (one that has a campaign-level daily budget). UNITS: new_daily_budget is in the account currency's MINOR units (cents), the unit Meta stores and get_campaigns reports: 2000 = €20.00/day. This differs from create_campaign / create_adset, whose daily_budget is in MAJOR units (40 = €40/day). To avoid converting, pass new_daily_budget_major instead (20 = €20.00/day); pass one of the two. Refused on an ABO campaign (budgets on the ad sets), because Meta would convert it to CBO and delete every ad-set budget, and on a lifetime-budget campaign. Use this to scale winning campaigns or reduce spend on underperformers. Returns a before→after receipt and logs to the decision ledger (auditable via get_decision_history).

Scale a campaign's daily budget by a percentage. Example: +20% increase or -15% decrease. Automatically fetches the current budget, computes the new value, and applies it. Returns the old→new budget and logs to the decision ledger (auditable via get_decision_history).

Re-enable a paused ad by setting its status to ACTIVE. Use this to reactivate ads that were previously paused.

Re-enable a paused ad set.

Re-enable multiple paused ads at once. Use to reactivate a batch of ads.

Create an automation rule from natural language. Converts the user's intent into a structured rule with conditions and actions. Example: 'pause any ad with CTR below 0.3% after spending €10' → rule with metric=ctr, operator=<, value=0.3, action=pause_ad. WORKFLOW: (1) compile the user's intent, (2) run backtest_rule with the same condition and SHOW the user what it would have done over the last 30 days, (3) create only after they confirm. Default mode is 'suggest' — the rule proposes and the user approves each action; only set mode='auto' when the user explicitly asks for automatic execution.

The ad pause recommendations from strategic analysis, read-only, for review before anything stops delivering. After you confirm, each ad is paused with pause_ad.

Pause a specific ad by setting its status to PAUSED. Use this when an ad has high frequency, poor CTR, or excessive CPL. Every pause is logged to the decision ledger with a before-snapshot (auditable via get_decision_history) and is reversible with enable_ad.

Pause an ad set. All ads within it will stop delivering.

Pause multiple ads at once. Use when you need to pause all bleeders in a campaign or a list of specific ads. More efficient than calling pause_ad multiple times. Each pause is individually logged to the decision ledger with a before-snapshot; one ad failing never aborts the rest.

Set a campaign's status to ACTIVE (launch it) or PAUSED (stop it). This is the final step of an MCP build chain: create_campaign → create_adset → create_lead_ads → enable_adset + bulk_enable_ads → set_campaign_status ACTIVE. Without this a campaign created through the MCP stays PAUSED forever. Verified by a post-write read-back. Note ACTIVE on the campaign alone does not deliver — the ad set and ads must be ACTIVE too.

Archive a campaign (status ARCHIVED): the campaign and every ad set and ad in it stop delivering. Use it to clean up test or finished objects, including ones created with create_campaign / create_adset / create_website_ad. The current status and budget are saved to the decision ledger. An object that is already ARCHIVED is left alone. Not reversible through AutoAdy: an archived object stays readable (history, results) but does not run again. To reuse it, duplicate it first. Always asks for a typed confirmation.

Archive a ad set (status ARCHIVED): every ad in it stops delivering. Use it to clean up test or finished objects, including ones created with create_campaign / create_adset / create_website_ad. The current status and budget are saved to the decision ledger. An object that is already ARCHIVED is left alone. Not reversible through AutoAdy: an archived object stays readable (history, results) but does not run again. To reuse it, duplicate it first. Always asks for a typed confirmation.

Archive a ad (status ARCHIVED): it stops delivering. Use it to clean up test or finished objects, including ones created with create_campaign / create_adset / create_website_ad. The current status and budget are saved to the decision ledger. An object that is already ARCHIVED is left alone. Not reversible through AutoAdy: an archived object stays readable (history, results) but does not run again. To reuse it, duplicate it first. Always asks for a typed confirmation.

Create a NEW Meta campaign — the top-level container an ad set and ads live in. Created PAUSED (never spends until you launch it). Default objective is OUTCOME_LEADS (lead-gen / instant-form). This is the top of the build chain: create_campaign → create_adset (copy a proven ad set's targeting via target_campaign_id) → create_lead_ads (load creatives). Budget: OMIT daily_budget for ABO (budget set per ad set — the common lead-gen case), or PASS daily_budget for CBO (one budget shared across the campaign's ad sets).

Create a NEW EMPTY ad set (no ads) in two possible ways. (A) COPY MODE — pass source_adset_id to clone a proven ad set's targeting, optimization goal, budget and promoted object; best when you want to reuse exact existing targeting such as a verified city/radius setup. (B) BUILD MODE — omit source_adset_id and pass campaign_id + targeting params to build one from scratch: countries and/or cities with radius, age, genders, locales, placements, optimization goal, and an ABO daily_budget with an optional bid cap. Created PAUSED. Follow with create_lead_ads (instant-form ad sets) or create_website_ads (OFFSITE_CONVERSIONS/website ad sets) to load creatives. If copy mode fails because the source uses a bid cap or hits Meta's copy limits, use build mode, or duplicate_adset with include_ads: false.

Duplicate an ad to a different ad set. Use to copy winning creatives into new audiences or test campaigns.

Duplicate an ad set to the same or different campaign, preserving its EXACT targeting (city/radius geo, exclusions), bid strategy and bid cap — things create_adset cannot always reproduce. New ad set is created PAUSED. By default the ads come too; pass include_ads: false to copy the ad set SHELL ONLY and then load your own creatives with create_lead_ads (instant-form ad sets) or create_website_ads (website and traffic ad sets). Use include_ads: false when the source has several ads — Meta rejects a deep copy of more than a couple of objects ('fewer than 3 objects per copy'), and a shell copy has no such limit. The receipt states the copy's budget as read back from Meta; copying an ad set from a campaign-budget (CBO) campaign into one without a campaign budget needs daily_budget.

Update specific targeting fields on a LIVE ad set without touching the rest: countries (narrow a broad/Europe-wide set to an exact country list), location_types (['home'] = residents only — removes travelers/visitors; ['home','recent'] = Meta's default), Advantage+ audience expansion on/off, age range, and placements. Read-modify-write: the current targeting is fetched first and ONLY the passed fields change — everything else is preserved exactly. The previous targeting is saved to the decision ledger for rollback. Placement notes: Meta re-validates the WHOLE targeting object on any edit, so a legacy manual placement set it now considers invalid (e.g. 'Instagram Explore home without Instagram Explore') blocks EVERY update — fix it in the same call by passing instagram_positions with the missing position added, or advantage_placements: true to switch to automatic. Note: editing targeting on a delivering ad set resets its learning phase.

Create a NEW Instant Lead Form — the form that opens when someone taps a lead ad. Returns a form_id you then attach with create_lead_ad / create_lead_ads (lead_gen_form_id). Use CUSTOM multiple-choice questions to gate for intent (a 'hard-gate' qualifying form) and higher_intent:true to add Meta's extra review step before submit. Without this tool a new ad set can only inherit an OLD form from an existing ad — this is how you launch with a fresh, purpose-built form.

Create a lead-form (instant form) ad from a CUSTOM image URL in an existing ad set. Uploads the image and attaches the lead form + page so it is a real instant-form ad (not a link ad). Created PAUSED. If lead_gen_form_id is omitted it inherits the form from an existing ad in the ad set. Use after create_adset to load your own creatives with matched copy.

Batch of create_lead_ad — create MANY lead-form ads (each a custom image + its matched copy) in ONE existing ad set in a single call. Up to 15. Created PAUSED, throttled to protect the account from Meta rate limits, and resilient: one ad failing never kills the rest. Returns which succeeded/failed. Use to load a whole creative batch at once.

Create a WEBSITE (link) ad in an existing ad set — it clicks through to a landing page URL instead of opening an instant form. This is the ad type for pixel/conversion ad sets (create_adset with OFFSITE_CONVERSIONS + pixel_id). The creative comes from EITHER image_url (uploads a new image) OR source_ad_id (reuses a winning ad's EXACT image by hash, no re-upload, and inherits its primary text/headline/description/CTA unless you override them — the way to carry proven creatives into a new website campaign). Created PAUSED. Instant-form ad sets (LEAD_GENERATION) are refused with guidance instead of a cryptic Meta error — use create_lead_ad for those.

Batch of create_website_ad — create MANY website (link) ads in ONE existing ad set in a single call. Up to 15. Each ad brings its own creative: image_url to upload, or source_ad_id to reuse a winning ad's exact image + copy (override any field per ad). The top-level link_url and url_tags apply to every ad unless overridden per ad. Created PAUSED, throttled against Meta rate limits, and resilient — one ad failing never kills the rest. This is the tool for 'rebuild the winners as a website-conversion campaign'.

Create a WEBSITE custom audience from a pixel rule — entirely via API, no Business Manager. Two rule kinds: url_contains ("/booking" → retargeting audience of page visitors) or event_name ("Lead" or a custom name like "lead_booked" → CONVERTERS audience, the ideal lookalike seed and exclusion list). Default: all website visitors. If Meta requires the one-time Custom Audience Terms acceptance, the error returns the exact one-click link — the only step the API can't do. The audience fills as the pixel fires (needs ~100+ people to serve or seed a lookalike).

Create a LOOKALIKE audience from an existing custom audience — Meta finds people similar to the seed (e.g. a converters audience from create_custom_audience). ratio 0.01–0.10 (1% = tightest match, start there), one 2-letter country per call. The source must belong to this account and hold 100+ people in the target country; Meta populates the lookalike over ~1–24h. Target it with create_adset (custom_audience_ids).

Start AI creative generation. Generates a brand-aware ad image. BEFORE calling this tool, you MUST confirm with the user: (1) what language the ad text should be in, (2) any specific message or hook to use. Only call this tool ONCE per user request — never retry.

Full pipeline: extract brand DNA from a URL and generate up to 50 ad creative variants across multiple angles, hooks, and visual styles. Returns task IDs for image generation. Best for: new brands, product launches, initial creative library. Paid plan required.

Take a winning ad and generate 10+ variations preserving the winning DNA but rotating visual style, hook, and messaging angle. Provide ad_id or image_url. Best for: scaling proven creatives, creative refresh.

Scientific A/B test: specify angles × hooks × visual styles to generate all combinations (max 27). Each combo produces a unique ad creative for testing.

Enable the autonomous creative loop: auto-detects fatiguing ads, generates replacements, optionally auto-launches winners. Paid plan required.

Extract brand DNA from a website URL: visual identity, voice profile, target avatar, winning patterns, unique mechanism. Foundation for all creative generation. Paid plan required.

Analyze which demographics convert best and suggest matching avatar descriptions + settings for creative generation. Saves audience demographics to brand memory.

Generate placement-specific creatives: Feed (4:5), Reels (9:16), Stories (9:16), Right Column (1:1). Each gets content designed for its consumption pattern.

Generate funnel-stage-specific creatives: Cold (awareness), Warm (consideration), Hot (conversion), Buyer (retention). Each stage gets psychologically matched content.

Analyze recently failed/paused ads to extract anti-patterns. Learns what NOT to do and saves patterns to Creative DNA for future avoidance.

Turn customer testimonial quotes into social proof ad creatives. Provide an array of quote strings and get branded testimonial ads.

Create a multi-week content calendar with themed weeks and angle rotations. Saves the calendar to brand memory for scheduled execution.

Maximum-scale product launch: generates 50 ads across 6 messaging angles from a single URL. Uses all available angles for maximum coverage.

Generate creatives optimized for multiple platforms: Meta Feed, Meta Reels, TikTok, Pinterest, Website. Each platform gets native aspect ratios and style.

Check the status of a running workflow (url_to_ads, multiply_winner, etc.).

Analyze a competitor's ad strategy using the Meta Ad Library. Provide a competitor name or page ID to see their messaging themes, creative styles, CTA patterns, and opportunities to differentiate. Needs a Growth or Agency plan.

Analyze competitor Meta ads, reverse-engineer their best creatives into templates, then generate your own branded versions using your brand DNA. Agency plan required.

Search the public Meta Ad Library by page id or search terms in one market and save what it returns as a competitor snapshot. Reaches outside this account; needs the workspace's Meta connection. Results are scoped to the authenticated workspace and account.

List the tracked competitor board: saved snapshots and change alerts, with filters and cursor paging. Read-only. Results are scoped to the authenticated workspace and account.

Compare recent competitor observations against this account's own creative history and return the concepts that are new rather than similar or duplicate, with citations. Read-only. Results are scoped to the authenticated workspace and account.

Start monitoring a tracked competitor source for changes on a cadence in seconds. Alerts land on the competitor board; this call does not capture anything itself. Results are scoped to the authenticated workspace and account.

Pause a competitor monitoring routine by routine_id. It stops checking for changes until resumed. It does not pause any ad. Results are scoped to the authenticated workspace and account.

Resume a paused competitor monitoring routine by routine_id, so it checks for changes on its cadence again. Results are scoped to the authenticated workspace and account.

Record one refresh of an active competitor monitoring routine: the content hash observed, when, and with what coverage. Raises a change alert when the content is new, changed or removed. Idempotent on idempotency_key. Results are scoped to the authenticated workspace and account.

Dismiss a competitor change alert by alert_id, or restore a dismissed one. Idempotent on idempotency_key. Results are scoped to the authenticated workspace and account.

Create a draft creative brief from research: facts, hypothesis, cited snapshots and observations, approved offers and claims, and the image and video direction. The draft cannot generate anything until a reviewer approves it with approve_creative_brief. Results are scoped to the authenticated workspace and account.

Read one creative brief version by version_id, with the evidence it cites. Read-only; use it before editing, approving or generating. Results are scoped to the authenticated workspace and account.

Change a creative brief by version_id with a patch of the fields to replace. The edit is saved as a new version with no approval of its own, so it must be approved before it can generate. Results are scoped to the authenticated workspace and account.

Approve a creative brief version so it can generate media. Records the approving user, and is refused if an offer or claim the brief cites is no longer approved or has changed. Results are scoped to the authenticated workspace and account.

Withdraw approval from a creative brief version, with an optional reason. Generation from that version is refused until it is approved again. Results are scoped to the authenticated workspace and account.

Start image or video generation from an approved creative brief. Spends AutoAdy credits and is idempotent on idempotency_key, so a retry with the same key does not start a second run. Poll with get_generation_status. Results are scoped to the authenticated workspace and account.

Read the state of a generation run by run_id: queued, running, succeeded, failed, cancelled or uncertain, with progress. Read-only; call it after generate_from_brief or resume_generation. Results are scoped to the authenticated workspace and account.

Resume an interrupted generation run by its original idempotency_key. Continues the same run instead of starting a new one. Results are scoped to the authenticated workspace and account.

Record a reviewer's approval of one generated artifact by artifact_id, with an optional reason. Nothing is attached to a live ad by this call. Results are scoped to the authenticated workspace and account.

Record a reviewer's rejection of one generated artifact by artifact_id, with an optional reason. Nothing in the ad account changes. Results are scoped to the authenticated workspace and account.

Build a preview of a Meta creative test plan: scope, brief, approved artifacts, control and variant, objective, metric, allocation, dates, budget, and success and kill rules. Returns a preview hash and confirmation token; nothing is created in Meta. Results are scoped to the authenticated workspace and account.

Approve a previewed Meta test plan by sending back the preview with its previewHash and confirmationToken. Refused if the preview changed since it was built. Results are scoped to the authenticated workspace and account.

Register a creative experiment: control and variant references, objective, metric, audience, source, provider, a settled measurement window and attribution. Idempotent on idempotencyKey. Record outcomes with record_experiment_result. Results are scoped to the authenticated workspace and account.

Record the measured result for one side (control or variant) of an experiment: sample size, spend, events, numerator and denominator, and when the data was fresh. Idempotent on idempotencyKey. Results are scoped to the authenticated workspace and account.

Attach a written interpretation to an experiment's results, optionally naming the hypothesis. Causal or unsupported performance claims are refused. Stored with the author; the numbers do not change. Results are scoped to the authenticated workspace and account.

Assess an experiment by experiment_id: whether the recorded control and variant results are sufficient evidence to learn from, and what is missing if not. Read-only. Results are scoped to the authenticated workspace and account.

Propose a learning from an experiment, with the text and the rationale behind it. It is saved as a pending candidate that a reviewer approves or rejects with review_learning_memory. Results are scoped to the authenticated workspace and account.

List experiments, results, interpretations and learning candidates for this account, or for one experiment when experiment_id is given. Read-only. Results are scoped to the authenticated workspace and account.

Approve or reject a pending learning candidate by candidateId, with an optional reason. Records the reviewer's decision; nothing in the ad account changes. Results are scoped to the authenticated workspace and account.

List read-only CRM and analytics connections with account, coverage, and freshness.

Read one CRM or analytics connection status and permissions.

Run a bounded read-only query against a connected CRM or analytics source, including CRM stage history.

Build a CRM pipeline report from one to four imported integration connections. Returns source evidence, freshness, partial reasons, and currency-separated exact amounts; it does not infer ROAS or historical conversion joins.