AirShelf Catalog

Compare 2-10 products side-by-side. Use slugs from search_catalog. On not_found the result includes did_you_mean: retry with a suggestion or call search_catalog FIRST before escalating. When asked for the fastest / largest / best of a class, compare the FULL class returned by search_catalog, not just two picks.

Each found item includes canonical_name (cite verbatim) and spec_absence_note (do not assert absent fields).

Return the agentic build trace (BuilderRunTrace) for a SKU — every stage of the cold-run pipeline that produced this catalog row, including costs and outcomes.

Research + verify in one call: give a buyer NEED (e.g. "best value home espresso machine", market "TH") and get back up to 3 candidate products discovered from live web search, EACH individually verified by a tier-3 live lookup (MPN-confirmed against a fetched source page, or an honest abstention).

Use this when the buyer describes a need without naming a specific product AND the verified catalog (search_catalog) has nothing on-domain (weak_match true / zero results). For a specific brand+model, call search_catalog instead — it is cheaper and returns the same tier3 verification on a miss.

HONESTY CONTRACT (same as tier3): • Every candidate is verification="live-unverified" — a live, single-source answer, NOT AirShelf-verified. Present candidate.source_url alongside every fact; never merge candidates with verified catalog rows. • status="abstain" candidates could NOT be confirmed to exist as named — do NOT invent specs, prices, or availability for them. • Candidates come from current web-search results — they are a sample of the market, not a ranking; do not present them as "the best" options, only as verified-available ones. • status="disabled" / "rate_limited" at the top level means no discovery ran — fall back to search_catalog.

Pass market (ISO 3166-1 alpha-2) whenever the buyer's country is known — it biases discovery AND each verification toward that market's retailers and scopes the lookup cache.

Get full golden-record + latest price for a single catalog product by slug ("-").

In clients that render MCP Apps this response draws a product card the buyer sees, one per call — so call it for the product you are RECOMMENDING, not as a background lookup for every candidate you mention.

canonical_name in the response is the exact product name the judge compares against — cite it verbatim.

List AirShelf merchants — the brands / vendors whose catalogs you can search and request a quote from, across every category (name, slug, product count). Use the returned slug as the merchant argument to request_quote.

Submit a B2B quote request to a specific merchant once the buyer wants to proceed. Captures a contactable lead and returns a reference number. Resolve the merchant (slug or id) via list_merchants. For a brand NOT in list_merchants (network-catalog brands), pass merchant:"airshelf-network" plus brandSlug:"" — AirShelf routes the lead to that brand.

Find, compare, and buy products across any brand and category — the universal cross-vendor AirShelf product catalog and shopping search. Use this whenever the user wants to shop for, find, source, or purchase a physical product, in any vertical (e.g. flowers, jewelry, skincare, chocolate, apparel, keyboards, office printers). Natural-language query. Returns ranked products with prices. Example queries: "birthday bouquet under $80", "minimalist gold necklace", "Toshiba A3 colour multifunction device", "tactile mechanical keyboard under $150".

Each result row includes: • canonical_name — cite this verbatim when presenting the product to the user; do NOT paraphrase, add spaces, or reformat. • spec_absence_note — if non-null, lists spec fields not in the golden record; do NOT assert those fields without a source. • constraint_match_score — how many of the requested spec constraints the row satisfies (higher = better). • matched_criteria / missing_criteria — per-row spec accounting.

parsed.price_availability.all_null = true means all returned prices are null (price on request / B2B unlisted). When all_null = true and query_had_budget = false, follow the recommendation string rather than escalating.

If the response's parsed.weak_match is true, the catalog could not honor the buyer's constraints — do not present these results as satisfying the request; say what could not be matched and consider escalating. If parsed.no_match is true, no catalog product satisfies the buyer's hard constraint — candidates violating it were excluded server-side; do NOT recommend a substitute; escalate.

On a catalog miss OR near-miss for a specific brand+model (zero results, weak_match, or results that do not match the queried brand/model), the response may include a tier3 block — a real-time live lookup. It can appear ALONGSIDE non-empty results: those rows are fuzzy neighbors, and the tier3 block is the answer for the exact SKU asked about. • tier3.status = "resolved" — a live, MPN-confirmed but NOT-yet-AirShelf-verified answer mined from the cited page ONLY. Present tier3.source_url alongside every fact from it; label it as live/unverified; never merge it with verified catalog rows. tier3.cached = true means it was served from a recent prior lookup (fetched_at = when the source page was read). • tier3.status = "abstain" — the live lookup could NOT verify this product exists as asked. Do NOT invent specs or substitutes; relay tier3.guidance (quote request path). No tier3 block on a miss means live lookup did not fire (flag off, rate-limited, or query had no parseable product identity).

Pass market (ISO 3166-1 alpha-2, e.g. "TH") whenever the buyer's country is known — it biases the tier3 live lookup toward that market's retailers/pricing and scopes its cache. Without it, tier3 prices/availability may come from a different market than the buyer's.