compare_rvs
ChatGPTCompare 2-4 MatchRV units side by side against the buyer's constraints. Keeps unknown values explicit and returns deterministic comparison evidence rather than guessed specifications.
evaluate_tow_fit
ChatGPTEvaluate weight fit between a shopper-stated tow vehicle and 1-6 RVs, or (with trailer_weight_lbs) whether 1-6 motorhomes can tow a loaded trailer. Motorhome mode checks each limit separately — loaded coach vs GVWR first, then hitch tow rating, GCWR, tongue limit and axle loads — and never reports 'safe'; the best result is 'not yet confirmed' with the ratings that clear. This is screening guidance, not a towing-safety guarantee.
get_dealer_inquiry_status
ChatGPTRead-only status of an inquiry this shopper sent: delivered / queued_for_matchrv / failed / dry_run, plus the dealership's reply only if MatchRV has recorded one. Requires the inquiry_id and private status_token returned by send_dealer_inquiry; other shoppers' inquiries are never visible.
get_rv
ChatGPTGet the full canonical MatchRV record for one unit returned by search_rvs, including dealer, specs, price, features, source/provenance, freshness, and explicit unknowns. Motorhomes include a powertrainAndTowing section (what powers it / what it can carry / what it can tow), each value with its status. Never invent missing specifications or assign an engine family's top rating to a unit.
outfitter_discovery
ChatGPTReturns the next question in MatchRV's guided RV-fit interview and, once enough fit facts are known, the search constraints for search_rvs. It takes only RV-fit facts: RV type, make or model, use category, desired features, party size, tow vehicle (or what a motorhome will tow), length limit, shopping area and radius, and budget with payment method (plus down and monthly payment if financing). Conversation-only questions (time off, who's buying, a trade-in, planned trips) are tracked by step name in covered_steps; their answers are never sent to MatchRV. Every input is optional, and the shopper may decline any question (declined_questions); a covered or declined step is never asked again. This tool never asks for names, contact details, income or credit, stores nothing, and contacts no one. Returns: status (needs_discovery, ready_for_products, or answer_directly when request_kind is general_question); nextQuestion (exactly one, with its step and a short reason for asking, to ask close to word for word; the trade question carries a followUp for when they have one); knownSummary; and, when readyForProducts is true, searchConstraints for search_rvs. Stateless and read-only: pass the fit facts and covered_steps gathered so far on every call.
outfitter_guide
ChatGPTMatchRV's buyer education, distilled from its published guides and the Road to the Sale method. Topics: road_to_sale (how buying works, step by step), how_to_pay (the six ways: cash, difference, finance, discount, ad shopper, special finance; out-the-door math), owners_and_credit, campsite_setup, tow_basics, motorhome_vs_trailer, trailer_vs_fifth_wheel, families, full_time, used_buying, costs, financing. Use it for 'how does this work / which is better / how should I pay / how do I set up camp' questions BEFORE any inventory search: teach first, never pressure. Pass topic for an exact guide or query to search. Towing points defer to evaluate_tow_fit for the buyer's actual vehicle, so never quote a model-level tow table.
outfitter_memory
ChatGPTRead what MatchRV remembers for THIS shopper (facts they taught via teach_outfitter) plus MatchRV-curated reference facts, each with confirmation counts and sources. Read-only. Use it to check for a remembered rating before asking again.
prepare_dealer_inquiry
ChatGPTStep 1 of contacting a dealership about one RV. Use only when the shopper asks to contact the dealer. Takes unit_id (from search_rvs/get_rv), the shopper's name, email and/or phone, and their message. Validates them and returns the exact preview — dealership name and location, RV title, VIN, stock number, price, the shopper's contact details, the message, and how MatchRV will deliver it — plus a confirmation_token valid for 30 minutes. NOTHING IS SENT. Show the whole preview to the shopper and ask them to approve it before calling send_dealer_inquiry.
search_rvs
ChatGPTSearch nationwide US MatchRV dealer inventory (Road to the Sale step 3). Pass make (e.g. "Tiffin") and model (series, e.g. "Allegro RED", "Phaeton", "Allegro Bus", "Allegro Open Road", "Wayfarer") whenever the shopper names them. They are hard filters, and the brand is reconciled from the listing title, so mislabelled dealer records are handled. model accepts several values; exclude_makes/exclude_models drop what they ruled out. Pass fuel_type ONLY when the shopper explicitly asks for gas or diesel. "Motorhome" means Class A, B or C. Inputs are RV-fit facts only (type, make/model, use category, desired features, party size, tow vehicle or towed vehicle, length limit, area and radius, budget with payment method), plus covered_steps / declined_questions: the interview steps already asked or declined, by name only. Conversation-only answers (time off, who's buying, a trade-in, planned trips) are never sent. Every input is optional. This tool never asks for names, contact details, income or credit, stores nothing, and contacts no one; contacting a dealer is a separate step that needs the shopper's explicit approval. ASK, DON'T ASSUME: pass only answers the shopper actually gave, and pass every one you have on each call. Until every interview step is covered (or declined), it returns status=needs_discovery with results=[] and exactly one nextQuestion (with its step and reason for asking); ask that one question and wait. A named make/model does not skip the questions. Skip only when the shopper literally asks: pass shopper_requested_results_now=true with their words in shopper_words; it shows results and returns resumeQuestions. Results are a shortlist of the 3 best fits from different model lines (limit up to 10 only when the shopper asks for more); never quote a total match count or how many of a model exist. Clarify ambiguous shopper needs, confirm approved flexibility, and preserve every hard constraint. State-wide location example: Florida; city-radius example: Tampa, FL. Pass location (city like "Fife, WA") to hard-filter by distance — never invent local matches from FL/AZ when the shopper asked for Washington. Also accept query (short search keywords like "travel trailer sleeps 8", not the shopper's message), sleeps_min, radius_miles, and structured constraints (priceMaxUsd, rvTypes, sleepsMin, lengthMinFt/lengthMaxFt, fuelType, horsepowerMin, torqueMinLbFt, trailerWeightLbs, trailerTongueLbs, coachLoadedWeightLbs). Use horsepowerMin/torqueMinLbFt only when the shopper states a minimum; use horsepowerPreferred/torquePreferredLbFt for "preferred". A minimum is never relaxed: an unknown rating is 'needs verification', never a match, and a lower rating is excluded. Motorhome rows include fuel, hp, torqueLbFt, hitchTowRatingLbs and gcwrLbs (null = not published). If coverage.no_local_matches, say there is no local inventory and offer to widen radius. sleepsConfirmed tells you whether capacity is dealer-published or inferred.
send_dealer_inquiry
ChatGPTStep 2: sends the inquiry previewed by prepare_dealer_inquiry. Consequential — it shares the shopper's name, contact details and message outside MatchRV. Call only after the shopper explicitly approved the exact preview, with confirmed=true. The content is locked to the preview (it cannot be changed here). Returns an inquiry_id, a private status_token and an honest status: delivered (emailed to a dealership that accepts MatchRV inquiries), queued_for_matchrv (MatchRV staff were alerted and will forward it; not yet at the dealer), or failed (tell the shopper to contact the dealership directly). Never retry automatically and never claim the dealer has replied.
teach_outfitter
ChatGPTRemember a fact the shopper confirmed, for THIS shopper only: their vehicle's exact tow rating or door-sticker payload (vehicle + tow_rating_lbs/payload_lbs), or a confirmed spec for one unit (unit_id + field + value; fields: sleeps, gvwrLbs, dryWeightLbs, hitchWeightLbs, freshWaterGal, bunkhouse). This is the only tool that writes memory. Facts are scoped to the shopper's ChatGPT identity / MCP session (or the returned session_key) and are never shown to other shoppers; they expire after 30 days of inactivity. Only teach facts the shopper actually confirmed — never guesses.