Find direct competitors running Facebook ads.

Use the connected Spytrend MCP to define the direct-competitor boundary.

Supported niche: home_services
Focus: Roof replacement contractors
Country (ISO-2): US
Lookback: 30 days of completed UTC days ending yesterday

End the window on the latest complete UTC day. Define a direct competitor at the company level from the supplied focus. Exclude directories, publishers, affiliates, lead networks, agencies, marketplaces and materially adjacent offers unless explicitly included by the focus. Derive and retain one or two short high-intent product/service phrases from the focus; never use the full focus sentence as a search query. Keep the category, focus, country and lookback fixed. Do not run discovery in this step.

Return no more than three concise bullets: confirmed scope and exact completed-date window; inclusion/exclusion boundary plus retained phrases; evidence gates. Do not show plan, quota, token balance, raw parameters or internal identifiers.

Stop after Step 1.

Cost guard: before each paid Spytrend call, run the free `estimate_cost` with the same tool and arguments, estimate the cost and stay within the user's stated budget; if the user has not stated a budget, tell them the estimate before the first paid call. If pricing consent is required, show the returned terms and stop until the customer explicitly agrees; never accept on the customer's behalf. If a required call does not fit the stated budget, make no paid call and return `Budget limit prevents required research`; omit an optional call that does not fit and disclose reduced coverage. When opening ads with `get_ad`, skip multilang ads (`is_multilang`) and never call `get_media` after `get_ad`. Retry once only when `retryable=true`, after `retry_after_seconds`. Read `params.ignored` after every search; an ignored required filter never counts as applied. If a search reports `pagination.total_status=restricted` (a free-plan archive window; `meta.plan_window_cutoff` names the boundary), tell the user the result covers only ads that started on or before that date, and never report a zero from it as "no such ads exist". Cost notices are not part of the final customer report.
Use the connected Spytrend MCP to discover candidate advertisers for the retained scope.

Run one primary heavy discovery route, chosen by the focus.

1. Focus set: run one `search_ads` with query = the first retained phrase, source Meta, categories=[home_services], countries=[US], status_today=active, the retained completed-day launch window, sort_by=date and limit=20. Run the second retained phrase the same way only when the first returns fewer than five plausible direct-fit advertisers. Do not use the niche-wide advertiser ranking when the focus is set; it rarely contains focus-specific businesses.
2. Focus blank: get the Meta advertiser ranking using the exact supported top-level category home_services, country US, dimension=advertisers, min_geo_share=0.5, min_category_share=0.3, sort_by=active_ads and limit=10. Do not add an AI subcategory; never infer one from customer text.

Deduplicate by advertiser identity and keep distinct pages separate unless evidence proves one company. Drop obvious info-guide, grant-assistance and lead-network pages. Retain up to eight candidates in returned order, each with its matching discovery ad when there is one. Discovery order is a signal, not proof of competitor fit, company size or performance.

If a discovery call returns retryable=true with retry_after_seconds, wait that delay and retry once identically. A final failure makes that configured route unavailable and the run Partial while unaffected routes continue. A completed zero is valid zero evidence; unavailable is never zero.

Return no more than eight concise bullets: route outcomes, delivered and deduplicated counts, and retained candidate names with source type. Retain exact IDs internally for Step 3 but never render full, shortened, masked or partial IDs. Do not show tables, calls, parameters, quota, token usage or rejected-row dumps.

Stop after Step 2.

Cost guard: before each paid Spytrend call, run the free `estimate_cost` with the same tool and arguments, estimate the cost and stay within the user's stated budget; if the user has not stated a budget, tell them the estimate before the first paid call. If pricing consent is required, show the returned terms and stop until the customer explicitly agrees; never accept on the customer's behalf. If a required call does not fit the stated budget, make no paid call and return `Budget limit prevents required research`; omit an optional call that does not fit and disclose reduced coverage. When opening ads with `get_ad`, skip multilang ads (`is_multilang`) and never call `get_media` after `get_ad`. Retry once only when `retryable=true`, after `retry_after_seconds`. Read `params.ignored` after every search; an ignored required filter never counts as applied. If a search reports `pagination.total_status=restricted` (a free-plan archive window; `meta.plan_window_cutoff` names the boundary), tell the user the result covers only ads that started on or before that date, and never report a zero from it as "no such ads exist". Cost notices are not part of the final customer report.
Use the connected Spytrend MCP to verify up to five retained candidate companies without rerunning discovery.

Take plausible direct-fit advertisers in returned order, up to five, so that 3–5 competitors can survive verification. For each, open its matching non-multilang (`is_multilang`) discovery ad once with `get_ad`. Only when discovery returned no matching ad, first run one candidate-specific `search_ads` with its exact advertiser ID, source Meta, countries=[US], active status, the retained completed-day window, date descending and `limit=3`, then open the best-matching non-multilang row once. When the ad qualifies, call its advertiser profile once for the company name and global active-ad count. Include a competitor only when the detailed ad is active, includes US, has a Facebook start date inside the exact window, explicitly promotes Roof replacement contractors, and retained evidence resolves company identity. Confirm the first-party landing relationship separately. Label advertiser-wide active-ad counts as global.

Exclude taxonomy-only matches, adjacent offers and unresolved company identities. Deduplicate verified pages into one company only when brand/domain evidence supports it. Do not open a second ad for the same candidate.

If a selected ad-detail or advertiser-detail call returns retryable=true with retry_after_seconds, wait that delay and retry once identically. A final call failure, ignored required filter or required budget block makes the run Partial. A completed detail that does not qualify the candidate and a missing optional landing field keep the run Complete; never replace a candidate after its ad is opened.

Retain full evidence internally. Return no more than eight concise bullets: verified, unresolved and excluded counts plus one short evidence summary per verified or useful unresolved company. Do not show raw IDs, calls, parameters, quota, token usage or detailed rejected-row dumps.

Stop after Step 3.

Using retained evidence only, write the final customer report without rerunning Spytrend.

Keep the complete answer under 380 words and table-free. Do not show raw IDs, category slugs, search phrases, calls, route names, quotas, token usage or detailed exclusions. Start with `# Competitors Running Meta Ads: Roof replacement contractors in ` followed by the full country name for US (write United States, not US; if the focus is blank, use the human-readable niche name, never its slug), then one summary line with verified competitor count, completed-day window and Complete/Partial.

Complete means every configured discovery and selected verification call completed successfully. A bounded sample or unresolved individual candidate does not make the run Partial. Use Partial only for a failed required route/call or unresolved required filter.

Show 3–5 compact numbered competitors when that many qualify; never pad. Include company, observed offer, active selected-country launch date, advertiser-wide active count labeled global, plain-text domain/landing confidence and exactly one review action.

Add `## Needs verification` only when populated, maximum one one-sentence entry. Add `## What stands out` with at most two bullets, `## Analyze next` with at most two companies and `## Research coverage` with at most two bullets. Do not claim that ranking order or global active counts measure company size, selected-country activity or performance.

End exactly: `Ad activity is a market signal, not proof of demand, spend, conversions, revenue, ROAS or profitability.`

Stop after Step 4.