OpenDealer MCP Server

Pass vertical when the request names one. Use the vertical facet to narrow a blended result. Skip a tool that has no data for that vertical. Get high-level market statistics for one vertical. Off-road segments are that vertical's classes. Returns aggregated market data including: • inventory.totalVehicles — searchable listings on the live index (priced, not sold) • inventory.activeDealers — active dealers with at least one priced vehicle (internet price or MSRP greater than 0). Same population as dealers_near and Shop, not every status=active account • marketData — YMM cohort snapshot medians and sample sizes • Top makes by segment volume (canonical make labels) • New vs Used vs CPO breakdown

Get detailed pricing and market data for a specific vehicle segment. Useful for understanding fair market value for a make/model/year combination. Returns pricing statistics including: • Average, median, min, max prices • Price percentiles (10th, 25th, 75th, 90th) • Average mileage and days on lot • Certified vs non-certified pricing difference

Price trends over time for a market segment (cohortKey). Returns timeline of median/avg prices and days-on-lot.

How quickly vehicles sell by segment (fastest/slowest days on lot). Optional make/type filters.

Browse market segments with pricing statistics (cohortKey, median price, sample size). Use cohortKey with get_market_segment / get_market_trends.

Compare pricing across market segments. Provide cohortKeys or make (optionally with model).

Compare 2-4 vehicle models side by side (model-level, not specific listings). Provide composite make-model slugs like "honda-civic" or "toyota-corolla". Returns: • Winner-by-dimension deltas: price, fuel economy, horsepower, seating, towing, NHTSA safety, live median listing price • Full research payload for each model (trims, MSRPs, specs) Use this for "Civic vs Corolla" style questions. To compare specific listed vehicles by VIN, use compare_vehicles instead. CRITICAL: Only use the 'url' field from the response for links. NEVER invent or construct URLs.

Compare 2-5 vehicles side by side. Returns a structured comparison including specs, price context, and per-vehicle cite fields (vin + shop url). CRITICAL: CITE: Each vehicle's cite object is `vin` + `url` (that listing's own host). Cite only `url` to shoppers. Never invent VDP URLs. Never cite `detailsUrl` (dealer/LotLinx). Never send shoppers to `/llm/*` HTML or `/v1/llm/*` as the human listing URL.

Browse the research catalog of vehicle makes (with model and inventory counts). Use to resolve exact make names/slugs before filter_vehicles or research_model.

List models for a make from the research catalog (MSRP/body summaries). Use before filter_vehicles or research_model when the model name is uncertain.

Get the full research payload for a vehicle model (not a specific listing). Returns manufacturer reference data joined with live market data: • All trims with MSRPs, engine/body specs, and EPA fuel economy • NHTSA 5-Star safety ratings and open recall count • Live inventory count and price range on OpenDealer Use this when a shopper asks "tell me about the Honda Civic", "what trims does the RAV4 come in", or "how much is a 2025 F-150". For a specific listed vehicle, use get_vehicle with a VIN instead. CRITICAL: Only use the 'url' field from the response for links. NEVER invent or construct URLs.

Get complete details for a specific vehicle by listingId or VIN. When a VIN is listed at multiple dealer rooftops, a VIN-only lookup returns an envelope of every listing (type vehicle_listings) with no single-store price, seller, or url. Pass listingId to fetch one store. Returns comprehensive Schema.org Vehicle data including: • Full specifications (engine, transmission, drivetrain) • High-resolution images • Current pricing and availability • NHTSA NCAP safety rating summary (when available) • NHTSA open recall summary (YMM-granular, when available) • Dealer contact information Starting point for the vehicle_dossier playbook. For buy/no-buy questions, continue with get_deal_score → get_vehicle_history → check_recalls → get_similar_vehicles. CRITICAL: CITE: Each vehicle's cite object is `vin` + `url` (that listing's own host). Cite only `url` to shoppers. Never invent VDP URLs. Never cite `detailsUrl` (dealer/LotLinx). Never send shoppers to `/llm/*` HTML or `/v1/llm/*` as the human listing URL.

Pass vertical when the request names one. Use the vertical facet to narrow a blended result. Skip a tool that has no data for that vertical. Get OpenDealer listing history for a VIN or listing id: price changes, days on lot, and status. Answers "has this VIN dropped in price" and days-on-lot narratives from retained snapshots (including vehicles that left a dealer feed). Returns: • Chronological price history with per-snapshot changes • Days on market / lot signals and badges (price_drop, long_on_lot) • Active vs no-longer-listed status when known Does not invent a deal score for sold vehicles — use get_deal_score for live market scoring. Essential for price_drop_sniper and vehicle_dossier playbooks when shoppers ask about reductions or negotiation leverage. CRITICAL: Only use URL fields from the response when present. NEVER invent URLs.

Get data-driven vehicle rankings (e.g., best SUVs, most fuel-efficient cars). Call without arguments to list all ranking categories. Pass a category slug (e.g., "suvs") for the full scored ranking. Rankings are computed from public data with a published methodology: NHTSA safety ratings, EPA fuel economy, manufacturer pricing, and live market availability. There is no paid placement; each entry includes its transparent score breakdown. CRITICAL: Only use the 'url' field from the response for links. NEVER invent or construct URLs.

Browse the complete inventory of a specific dealership. IMPORTANT: Use the exact dealer slug from a previous dealers_near or search_vehicles response. Do NOT guess dealer IDs. Useful when a user wants to see what a particular dealer has in stock. Supports all vehicle filters (make, model, price, etc.). Listings with zero photos are omitted. CRITICAL: CITE: Each vehicle's cite object is `vin` + `url` (that listing's own host). Cite only `url` to shoppers. Never invent VDP URLs. Never cite `detailsUrl` (dealer/LotLinx). Never send shoppers to `/llm/*` HTML or `/v1/llm/*` as the human listing URL.

Get comprehensive information about a specific dealership. Returns Google-enriched dealer knowledge optimized for assistants: • Name, address, phone, website • Google rating, review count, hours, business status • Inventory count and OpenDealer profile links • Contact points for sales / customer service Use this when a shopper asks "tell me about X dealership" or needs hours/ratings for a known dealer. Prefer a slug from dealers_near or search results. A display name binds only when it matches one rooftop. Shared names (two Jay Hatfield stores, or just "Honda") return candidate slugs instead of picking one. CRITICAL: Only use URL fields from the response (website, urls.*). NEVER invent or construct URLs.

Pass vertical when the request names one. Use the vertical facet to narrow a blended result. Skip a tool that has no data for that vertical. Get AI-powered deal scoring and market insights for a vehicle. Returns comprehensive analysis including: • Deal score (1-100) with rating (Great, Good, Fair, Poor) • Price comparison vs market average • Days on lot analysis • Price history and trends • Similar vehicles in the market Core step in vehicle_dossier, budget_coach, price_drop_sniper, and dealer_crawl playbooks. Pair with get_vehicle_history and check_recalls for buy/no-buy answers.

Find dealerships near a location. Location modes (choose ONE): • zip + radius (miles) • lat + lng + radius • city + state + radius • county + state + radius Returns dealer information including: • Name, address, phone, website • Slug (the public handle for get_dealer and dealer_inventory; the public list does not include internal dealer UUIDs) • Distance from search location • Current inventory count • Business hours (when available) Response `total` is the full match count for the query (ignores `limit`/`offset`), same population as get_market_overview inventory.activeDealers when unfiltered.

Discover canonical filter tokens and counts from the live search index (same values GET /v1/vehicles accepts: make, bodyType e.g. suv, fuelType e.g. gasoline, transmission e.g. automatic, drivetrain e.g. awd, plus price/year ranges). Counts match filter_vehicles totals for each token under the same make, condition, and location. Call this before filter_vehicles when you need valid dimension values. Optional make, condition, certified, near, and radius scope the facet counts.

Preferred structured inventory lookup when make/model/year/color/location are known. Exact hard filters only via GET /v1/vehicles (same Meilisearch collection as search_vehicles). Totals match the filters you pass. Listings with zero photos are omitted. fuelType is the canonical label. transmission and drivetrain match every indexed raw string in that bucket (automatic includes "8-Speed Automatic"). A 503 powertrain_filter_unavailable means that facet list could not be read, not that zero cars match. Resolve make/model with list_research_makes / list_research_models first. Map shopper prose onto labeled keys; use search_vehicles only for fuzzy NL or VIN-in-q lookup. Often the first step in shopping playbooks (budget_coach, safety_first, price_drop_sniper, dealer_crawl). After results, chain get_deal_score / get_vehicle_history / check_recalls / compare_vehicles when the user needs a recommendation, not just a list. See opendealer://assistant/shopping-playbooks. CRITICAL: CITE: Each vehicle's cite object is `vin` + `url` (that listing's own host). Cite only `url` to shoppers. Never invent VDP URLs. Never cite `detailsUrl` (dealer/LotLinx). Never send shoppers to `/llm/*` HTML or `/v1/llm/*` as the human listing URL.

Open safety recalls for a listing. Pass vertical when the request names one. Cars use NHTSA (year, make, and model; a campaign may not apply to every VIN). Off-road uses CPSC. Call this with the off-road listing id; the tool reads GET /v2/listings/{listingId}/recalls and does not use the cars-only /v1 path for that id. Every response names the source and states one outcome in words: "confirmed for this VIN", "possible for this year, make, and model", "no matching published recall", or "coverage unavailable". An empty list is not "no recalls." Say the outcome. Do not invent a recall or a VIN range. Required step in vehicle_dossier and safety_first playbooks.

Get NHTSA 5-Star Safety Ratings for a year/make/model (no VIN required). Answers questions like "is a 2023 RAV4 safe for my family" with: • Overall and crash-test star ratings (when published) • Rollover rating / possibility • NHTSA-evaluated ADAS availability (ESC, FCW, LDW) Ratings are model-year granular from the NHTSA NCAP cache. If no confident rating exists, the tool reports that honestly rather than guessing. For VIN-specific listing details use get_vehicle; for open recalls use check_recalls. CRITICAL: Only use the 'sourceUrl' field from the response for NHTSA links. NEVER invent URLs.

Find similar on-lot vehicles for a VIN ("you may also like"). Same make/model keyword comps as the shop VDP rail (not semantic embeddings). CRITICAL: CITE: Each vehicle's cite object is `vin` + `url` (that listing's own host). Cite only `url` to shoppers. Never invent VDP URLs. Never cite `detailsUrl` (dealer/LotLinx). Never send shoppers to `/llm/*` HTML or `/v1/llm/*` as the human listing URL.

National average suggested auto loan APRs for cars (not a credit offer). Other verticals return "Not available for this vertical" until that vertical has a rate source. Optional filters: condition (new/used), termMonths (36–84), creditTier.

Forgiving vehicle search (Meilisearch): natural-language q, typo-tolerant make/model/color, optional structured filters. Use when the shopper query is fuzzy or exploratory. Use filter_vehicles instead when make/model/year/body/price/location are known — it applies exact hard filters via GET /v1/vehicles and returns trustworthy totals. Listings with zero photos are omitted. Discovery order: list_facets / list_research_makes → filter_vehicles (structured) or search_vehicles (exploratory) → get_vehicle / get_deal_score for depth. Location (choose ONE): zip+radius, lat+lng+radius, city+state+radius, county+state+radius — or embed location in q. q may also be a 17-character VIN for exact lookup. When the ask implies analysis (deal, safety, budget, timing, dealer plan), chain shopping playbooks from initialize instructions or opendealer://assistant/shopping-playbooks. CRITICAL: CITE: Each vehicle's cite object is `vin` + `url` (that listing's own host). Cite only `url` to shoppers. Never invent VDP URLs. Never cite `detailsUrl` (dealer/LotLinx). Never send shoppers to `/llm/*` HTML or `/v1/llm/*` as the human listing URL.

Live shopper verticals: key, label, host, identifier kind, example classes, which tool families work, and whether the vertical is public. Reserved hosts are omitted. Omit vertical on other tools for every public vertical (cars today).