create_catalogue_experiment
ChatGPTUse this when the user wants to plan a controlled product-title catalogue experiment or A/B test. Creates and locks a matched concurrent Control/Treatment experiment plan. It does not publish changes; launch is a separate confirmed action.
diagnose_account
ChatGPTPPC triage of the whole Google Ads account, live: campaigns not serving, limited, spending without conversions or bidding without signal; tracking blockers; money at risk; and the conversion actions each live campaign bids on. Use FIRST for account health, audits, 'which campaigns' questions. Read-only.
diagnose_campaign_delivery
ChatGPTWhy a campaign gets no or few impressions/clicks, live: status, serving and primary status reasons, billing, bidding vs conversion signal, budget, ad approval/policy, keywords (volume, Quality Score), schedule, impression share. Returns verdict + ranked findings + fix. Accepts a campaign name. Read-only.
diagnose_conversion_setup
ChatGPTWhich conversion actions a campaign (or the account) really optimizes for (biddable goals, primary actions, custom goal), whether they recorded conversions in 30 days, secondary actions that record but are ignored, and goal vs business-objective mismatch. Accepts a campaign name. Read-only.
execute_ad_schedule_replacement
ChatGPTUse this only after explicit user approval of the exact prepared preview to replace a Google Ads ad schedule. Executes only that operation and requires the matching request ID and preview hash.
execute_audience_target_update
ChatGPTUse this only after explicit user approval of the exact prepared preview to update Google Ads audience targets. Executes only that operation and requires the matching request ID and preview hash.
execute_campaign_budget_update
ChatGPTUse this only after explicit user approval of the exact prepared preview to update a Google Ads campaign budget. Executes only that operation and requires the matching request ID and preview hash.
execute_campaign_enable
ChatGPTUse this only after explicit user approval of the exact prepared preview to enable a Google Ads campaign. Executes only that operation and requires the matching request ID and preview hash.
execute_campaign_pause
ChatGPTUse this only after explicit user approval of the exact prepared preview to pause a Google Ads campaign. Executes only that operation and requires the matching request ID and preview hash.
execute_conversion_goal_alignment
ChatGPTUse this only after explicit user approval of the exact prepared preview to align Google Ads campaign conversion goals. Executes only that operation and requires the matching request ID and preview hash.
execute_keyword_status_change
ChatGPTUse this only after explicit user approval of the exact prepared preview to change a Google Ads keyword status. Executes only that operation and requires the matching request ID and preview hash.
execute_keywords_addition
ChatGPTUse this only after explicit user approval of the exact prepared preview to add Google Ads keywords. Executes only that operation and requires the matching request ID and preview hash.
execute_location_target_update
ChatGPTUse this only after explicit user approval of the exact prepared preview to update Google Ads location targets. Executes only that operation and requires the matching request ID and preview hash.
execute_negative_keywords_addition
ChatGPTUse this only after explicit user approval of the exact prepared preview to add Google Ads negative keywords. Executes only that operation and requires the matching request ID and preview hash.
execute_responsive_search_ad_replacement
ChatGPTUse this only after explicit user approval of the exact prepared preview to replace a Responsive Search Ad. Executes only that operation and requires the matching request ID and preview hash.
finalize_catalogue_experiment
ChatGPTUse this only after explicit user confirmation to finalize a completed catalogue/title experiment by rolling out a verified winning challenger or restoring the original state. rollout_challenger is allowed only for a valid positive causal result. Magnify verifies the final state in Google.
find_money_leaks
ChatGPTSTART HERE when the user connects Magnify or asks where they are wasting money, what is wrong, or for a quick audit. Returns the top 3 money leaks in the authorized Google Ads account for the last 30 complete days: search terms, products and campaigns that spent with zero recorded conversions, each with observed spend, clicks, a plain-language reason and one fix. Amounts are observed spend to review, NOT guaranteed savings; repeat the caveat. To fix a leak, call ask_magnifai with the leak_id: it prepares a preview that the owner approves in Magnify. After the leaks, offer quick_actions (Roast My Ads, 92-Check Audit, 30-Day Plan, Feed Optimization, Ad Copy Audit) as short options, not a long plan. If state is syncing, say the first sync is still running and offer the quick actions; never say there are no leaks. Read-only.
get_account_health
ChatGPTUse this when the user asks for a Google Ads audit, account health check, conversion/tracking risks, configuration problems, or setup opportunities. Returns Magnify's latest 92-point audit findings, including failing, opportunity, and unknown checks. Read-only and evidence-based.
get_account_overview
ChatGPTUse this when the user asks for Google Ads performance, spend, revenue, ROAS, CPA, CTR, clicks, impressions, or conversion trends for an authorized Magnify workspace. Reads persisted daily performance rows. In the dedicated OpenAI reviewer workspace these rows are synthetic SAMPLE data and intentionally do not require a live Google Ads connection; in normal workspaces they come from the connected account. Use before performance claims; it does not explain causality by itself.
get_action_status
ChatGPTUse this when the user asks whether an approved Magnify Google Ads change actually ran, succeeded, failed, or is still pending. Returns the latest execution and result state for a previously prepared action. Read-only.
get_business_context
ChatGPTFIRST CALL after selecting an authorized store or before an account-specific PPC strategy. Returns owner-confirmed goals, website, ICP, target markets and missing_fields. If context is missing, ask one short question instead of guessing; the owner may edit it by preview + confirmation or via Business Settings. Read-only.
get_catalog_summary
ChatGPTUse this when the user wants a high-level audit of their Shopping catalogue or product-feed state. Summarizes product lifecycle, quality issues, primary-feed attribution, unattributed products, and pending or unverified publication counts. Read-only.
get_discovery_summary
ChatGPTUse this when a customer has just connected Google Ads or asks what Magnify found first. Returns a ranked read-only projection of existing Account Health and persisted recommendations, including source IDs, freshness, uncertainty and review next steps. It does not claim verified savings, automatically refresh evidence, prepare proposals or execute ads changes.
get_experiment_report
ChatGPTUse this when the user asks for the evidence, design, matched products, publication verification, history, or result of a specific catalogue/title experiment. Returns an auditable report. Causal wording requires result.health.causal_claim_allowed=true.
get_experiment_status
ChatGPTUse this when the user asks for the current status of title tests, image tests, A/B tests, or AutoPilot experiments. Lists canonical experiment state without changing or launching experiments. Read-only.
get_experiments
ChatGPTUse this when the user asks about product-title tests, catalogue experiments, A/B test status, or observed experiment effects. Lists Magnify matched Control/Treatment catalogue experiments. Treat results as incremental or causal only when result.health.causal_claim_allowed=true.
get_feed_health
ChatGPTUse this when the user asks to audit a Google Merchant Center or Shopping feed, diagnose feed health, check sync/connection state, or understand feed quality and publication problems. Summarizes canonical feed sources, attribution, quality issues, product lifecycle, and publication state. Read-only.
get_feed_publication_status
ChatGPTUse this when the user asks whether optimized product titles or other feed values are actually live in Google Merchant Center. Returns submission and verification evidence from the canonical publication ledger. Optimization or saving alone is never treated as proof that a value is live.
get_feed_quality_issues
ChatGPTUse this when the user asks what is wrong with their Merchant Center/Shopping feed, including missing titles, descriptions, images, or disapproved products. Returns canonical feed-quality issue counts for the selected feed or store. Read-only.
get_feed_sources
ChatGPTUse this when the user asks which Merchant Center feeds are connected, needs the exact feed to analyze, or a feed scope is ambiguous. Lists canonical primary and supplemental feed sources, locale, datasource identity, sync state, and product attribution. Read-only.
get_image_opportunities
ChatGPTUse this when the user asks which Shopping products have missing primary images, failed lifestyle-image workflows, or no lifestyle image. Finds deterministic image opportunities only; it does not generate images, modify feeds, or spend credits.
get_label_status
ChatGPTUse this when the user asks about Merchant Center custom labels, product-label coverage, or Magnify label rules. Summarizes effective custom_label_0 through custom_label_4 values and scoped label rules. Google Ads performance is intentionally excluded. Read-only.
get_magnify_decisions
ChatGPTUse this when the user asks why Google Ads performance changed, what Magnify recommends, what should happen next, or which changes are safe to execute. Returns the persisted Decision Stack with evidence, confidence, risk, learning state, expected impact, guardrails, and execution readiness. Prefer this before strategic claims.
get_magnify_ppc_plan
ChatGPTPRIMARY TOOL when the user asks ChatGPT or Claude to work on, diagnose, prioritize, or improve an authorized Google Ads account as a whole. NOT for ad texts or creatives: for current headlines/descriptions/assets use get_pmax_diagnostics (Performance Max) or list_search_ad_resources + get_search_ad_creative_evidence (Search). Returns Magnify's evidence-bound PPC operating plan: observed account KPIs, ranked health/Decision Stack findings, campaign and search-term review candidates, source freshness, and concrete next steps. This is not a generic Google Ads data connection. Never invent savings, infer business goals or execute any changes. To apply an action, a separately entitled Magnify operator must create an exact signed preview, obtain independent owner approval in Magnify, execute through the Magnify Action Router and verify Google Ads readback.
get_opportunities
ChatGPTUse this when the user asks what to improve, optimize, fix, or prioritize next in their Google Ads account. Returns persisted Magnify recommendations with evidence, confidence, expected impact, and guardrails where available. Recommendations are not automatically executed.
get_pending_feed_changes
ChatGPTUse this when the user asks whether feed changes are queued, stuck, publishing, submitted, verifying, failed, superseded, or waiting to reach Google. Reads the canonical publication ledger and recent changed fields. It does not publish anything.
get_performance
ChatGPTUse this when the user asks which campaigns are spending, growing, declining, wasting budget, or should be compared in an authorized Magnify workspace. Returns persisted campaign-level KPIs for the selected period, sorted by spend. In the dedicated OpenAI reviewer workspace these are synthetic SAMPLE campaign rows and do not require a live Google Ads connection. Read-only; use supporting evidence before recommending a change.
get_pmax_diagnostics
ChatGPTUse this when the user asks to audit, diagnose, troubleshoot, or improve Performance Max (PMax), including asset groups, ad strength, weak/disapproved creative assets, channel spend allocation, campaign KPIs, or product performance. Reads real connected Google Ads evidence via Magnify. Read-only; call this before saying PMax structure or channel data is unavailable.
get_product_details
ChatGPTUse this when the user asks about one specific Shopping/Merchant Center product: its title, optimization state, GMC status/issues, images, or whether an optimized value was published. Requires a known internal product UUID or merchant product ID. Read-only.
get_product_experiment_plan
ChatGPTRead-only feasibility and immutable plan preview for exact product UUIDs. Uses the same domain planner as the UI. Never creates, publishes, or promises a winner. Catalogue cluster designs may be unavailable. Requires authenticated actor context.
get_product_opportunities
ChatGPTUse this when the user asks which Shopping products need attention, have weak or missing titles, are disapproved, have GMC issues, image problems, or optimized content that is not published. Finds deterministic product opportunities from canonical catalogue/GMC fields; it does not use an AI score or change products.
get_roast_status
ChatGPTRefreshes a roast_my_ads run that is still syncing (the in-chat card calls this itself). Returns the same shape as roast_my_ads. Read-only.
get_search_ad_creative_evidence
ChatGPTRetrieve actual Google Ads responsive Search ad headlines, descriptions, pinning, final URLs and policy approval state by exact ad_group_ad_resource. Read-only, available on Free; never asks for Operator approval.
get_search_terms
ChatGPTUse this when the user asks where Google Ads spend is being wasted, which search terms cost money without conversions, or where negative-keyword opportunities may exist. Reads real search-term performance and can filter zero-conversion terms and minimum spend. It does not add negative keywords or change the account.
launch_catalogue_experiment
ChatGPTUse this only after the user explicitly confirms they want to start an already planned catalogue/title experiment. For Magnify publication this changes treatment product feed titles; for Lengow/external it returns the treatment publication manifest. Never launch without explicit confirmation.
list_search_ad_resources
ChatGPTDiscover live responsive Search ad resource names in an authorized Google Ads campaign (Search campaigns only; for Performance Max headlines use get_pmax_diagnostics). Read-only. Use this before get_search_ad_creative_evidence when an ad resource ID is unknown. Limited to 50 records and reports truncation.
list_stores
ChatGPTUse this when the user needs to choose or confirm which Magnify workspace to analyze. Lists only workspaces authorized for this connection and indicates whether a workspace is using reviewer SAMPLE data or a live connected account. Read-only. A sample workspace is intentionally valid for read tools and does not require a live Google Ads connection.
open_magnify_view
ChatGPTUse when the user wants to see, open or work in the Magnify app itself (dashboard, account health, opportunities, campaigns, keywords, insights, change history, reports, workspace settings, or tracking: the Google Tag Manager card where an owner connects GTM, chooses the container and allows changes) rather than a chat summary. Returns a link to that page of the signed-in Magnify web app for the authorized store. Always show it to the user as a clickable markdown link; Claude Desktop opens it beside the chat. The link contains no credentials: the user signs in to Magnify if needed, and the app enforces their own access. Read-only; never changes Google Ads.
prepare_ad_schedule_replacement
ChatGPTUse this only when a Magnify recommendation proposes to replace a Google Ads ad schedule. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_audience_target_update
ChatGPTUse this only when a Magnify recommendation proposes to update Google Ads audience targets. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_business_context_update
ChatGPTAuthenticated OWNER only. Prepare exact before/after business goal and optional website/ICP/market metadata in Magnify. Free and Operator support this PROFILE-ONLY preview (never changes Google Ads). Show it to the owner and WAIT for explicit confirmation in a subsequent chat message before confirmation.
prepare_campaign_budget_update
ChatGPTUse this only when a Magnify recommendation proposes to update a Google Ads campaign budget. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_campaign_enable
ChatGPTUse this only when a Magnify recommendation proposes to enable a Google Ads campaign. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_campaign_pause
ChatGPTUse this only when a Magnify recommendation proposes to pause a Google Ads campaign. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_conversion_goal_alignment
ChatGPTUse this only when a Magnify recommendation proposes to align Google Ads campaign conversion goals. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_keyword_status_change
ChatGPTUse this only when a Magnify recommendation proposes to change a Google Ads keyword status. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_keywords_addition
ChatGPTUse this only when a Magnify recommendation proposes to add Google Ads keywords. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_location_target_update
ChatGPTUse this only when a Magnify recommendation proposes to update Google Ads location targets. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_negative_keywords_addition
ChatGPTUse this only when a Magnify recommendation proposes to add Google Ads negative keywords. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
prepare_responsive_search_ad_replacement
ChatGPTUse this only when a Magnify recommendation proposes to replace a Responsive Search Ad. Creates the exact supervised preview and does not change Google Ads. Show the preview before asking for approval.
resolve_ads_entities
ChatGPTTurn a campaign or ad group name (partial, any case, with or without emoji) into exact Google Ads IDs with status and channel. Returns the resolved match or ranked candidates. Use instead of asking the owner for IDs. Read-only.
roast_my_ads
ChatGPTUse this when the user asks to roast, grade or audit their Google Ads, asks what's wrong with their Google Ads, where they waste spend, or why they aren't getting sales. Chef Brasa (Magnify's original chef character) grades the account A+ to F from the last 30 complete days of synced data and returns a headline, three roast lines, the top 3 leaks (each with a fix_prompt) and a rescue plan. Every number and term is checked against the account's data. Present the headline and the three lines exactly as returned. Do not rewrite or soften them. Amounts are spend to review, never guaranteed savings. To fix a leak, call ask_magnifai with that leak's fix_prompt: it prepares a preview the owner approves in Magnify. 'Turn up the heat' = call again with the next spice (free, same data). mode=reroast re-grades after 7 days or a verified fix. If status is syncing, too_fresh or limit_reached, relay its message. Read-only.
show_magnify_home
ChatGPTMAGNIFY HOME / COMMAND CENTER. Call FIRST whenever the user invokes Magnify in a new chat, asks what Magnify can do, asks for a menu or has just connected during onboarding. Returns visual entry points for six read-only workflows and authorized workspaces. Does not execute audits or Ads changes. If the user specifically requests one action, do that action directly instead. MCP cannot display itself before the host invokes Magnify.
stop_catalogue_experiment
ChatGPTUse this only after explicit user confirmation to stop an active catalogue/title experiment. Magnify restores the exact pre-experiment title state and verifies restoration in Google before cancellation; external adapters receive a final-action manifest. This is a destructive write action.
