bringatrailer.com logo
bringatrailer.com

find-cars

Installation

Adds this website's skill for your agents

 

Summary

Scan Bring a Trailer's live auctions for collectible car models whose current bid is below the model's historical BaT sale median, returning each candidate with market comps, days left, and reserve status. Read-only.

FIG. 01
FIG. 02
FIG. 03
FIG. 04
SKILL.md
233 lines

Find Below-Market Vintage Cars on Bring a Trailer

Purpose

Scan Bring a Trailer's live auctions for a set of collectible car models and surface the ones whose current bid is below that model's historical BaT sale median, along with the comps that justify the call. For each model it returns the market band (p25 / median / p75 from past BaT sales), every currently-running auction that matches the model, and per-listing signals — current bid, percent-of-median, days left, and no-reserve flag — so the caller can judge whether a low number is a real bargain or just an auction that hasn't been bid up yet. Read-only: it never bids, watches, comments, or logs in.

When to Use

  • "Find first-gen Broncos / Datsun 240Zs / Pagoda SLs currently going below their usual BaT price."
  • A collector or dealer monitoring live auctions for buying opportunities across a watchlist of models.
  • Building an alerting agent that flags undervalued lots ending soon.
  • Any time you'd otherwise hand-scrape BaT auction cards — the inline JSON feed is faster, complete, and structurally reliable.

Workflow

BaT is a WordPress site that embeds everything you need as inline <script> JSON, and both the live-auction page and each model page return that JSON to a plain HTTP GET — no login, no cookies, no anti-bot stealth. The optimal method is therefore fetch + parse, not scripted browsing. browse cloud fetch <url> returns a JSON envelope {statusCode, headers, contentType, encoding, content}; the HTML is in .content. --proxies is not required (bare GET returns 200 with the full blob); add it only if Cloudflare ever challenges you.

  1. Pull the live-auction feed (one fetch). GET https://bringatrailer.com/auctions/ → in .content, find the marker var auctionsCurrentInitialData = { and brace-match the object. .items is the complete live set (~960 lots). Each item has: title, current_bid (integer), current_bid_formatted, timestamp_end (epoch seconds), noreserve, year, url, and — critically — keyword_pages: an array of model-page IDs that is the precise join key back to a model.

  2. Resolve each target to its BaT model page. If you don't know the slug, hit the autocomplete endpoint: GET https://bringatrailer.com/wp-json/bringatrailer/1.0/data/autocomplete?term=<urlenc-name> (the param is term — q/query/keyword all return HTTP 400). .content is JSON whose items[] give {title, url}, e.g. term=pagoda → mercedes-benz/w113, term=365 → four Ferrari 365 variants.

  3. Get the model's market price + join IDs. GET https://bringatrailer.com/{make}/{model}/ → extract var auctionsCompletedInitialData = {.

    • Market band: prefer stats.sold[].amount — the model's entire BaT sale history in one array — and take the median (and p25/p75). Filter amounts >= $3000 to drop parts/wheels/seats lots that share the model tag.
    • Join key: base_filter.keyword_pages — the model's page ID(s).

  4. Match live auctions to each target. A live item belongs to a target iff its keyword_pages intersects the model page's base_filter.keyword_pages. This is far more reliable than title/keyword regex.

  5. Flag + rank. Mark matched live lots with current_bid < median as below-market candidates. Compute pct_of_median = current_bid / median * 100 and days_left = (timestamp_end*1000 - now) / 86400000. Always surface days_left and no_reserve — a bid at 20% of median with 6 days left is not a bargain, it just hasn't been bid up. The strongest signals are lots ending within ~1 day still well under median, and no-reserve lots (guaranteed to sell).

Resolving broad "family" models (generation drill-down)

Some names map to a family page with stats_enabled:false and an empty stats object — you cannot get a market price from it directly. These include Camaro, Bronco, Porsche 356, Porsche 911 GT3, and Mercedes Geländewagen. On the family page, each generation is listed in a data-items="[...]" HTML attribute (HTML-entity encoded — decode &quot;→", &#8211;→-, &amp;→&) as {id, title, url}. Pick the generation matching the requested era and fetch that page:

  • First-gen Camaro (1967-1969) → chevrolet/camaro-1967-1969
  • First-gen Bronco (1966-1977) → ford/bronco-u13-u14-u15-1966-1977
  • First-gen G-Wagen (W460) → mercedes-benz/w460-gelandewagen
  • Porsche 356 → per-variant pages (356a, 356b, 356c, 356-speedster, …)
  • 911 GT3 → per-generation pages (991-gt3, 992-gt3, 996-gt3, 997-gt3)

Browser fallback

If plain GETs ever get Cloudflare-challenged: open a remote session (browse open <url> --remote), let the page render, then read the same objects with browse eval 'JSON.stringify(window.auctionsCurrentInitialData)' and window.auctionsCompletedInitialData. Same fields, ~100x the cost — avoid unless forced.

Site-Specific Gotchas

  • browse cloud fetch returns an envelope, not raw HTML. Parse the outer JSON and read the page from .content. Inside .content the inline blobs use normal (unescaped) quotes — brace-match from the first { after var NAME = .
  • stats.sold is served intermittently (edge-cache variance). The same model URL may return a full stats.sold array on one fetch and omit it (empty stats) on the next — observed on mercedes-benz/w113 (absent 3× then 569 sales on the 4th) and ford/bronco-u13-… (1234 sales, then absent). Retry the fetch up to ~6× until stats.sold is present. If it never appears, fall back to the 24 items[] on completed page 1 (recent comps, lower confidence) and label the source.
  • Family pages have stats_enabled:false and empty stats. Don't try to price Camaro/Bronco/356/911-GT3/Geländewagen from the family page — drill into a generation page (see above). items_total is populated but the money data is not.
  • keyword_pages join pulls in parts & projects. Chassis, seats, tube-frame race cars, and engine-swap projects share a model's keyword_pages. The >= $3000 floor removes most parts, but inspect title before calling something a "below-market car" (e.g. a "$101 Camaro-bodied tube-frame track car" or a "$39,500 chassis for Bronco" are not cars).
  • Trim levels have no dedicated page. "911 GT3 Touring" is a trim, not a model — use the generation page (porsche/992-gt3) and filter title by /touring/i. Same pattern for other sub-trims.
  • Ambiguous model names resolve to several pages. term=365 returns 365 GT 2+2, 365 GTC, 365 GTC/4, and 365 GT4/400/412 — pick the intended one or price each.
  • Thin markets → low-confidence median. Ferrari 365 GTC/4 (~11 sales) and INEOS Grenadier (~24 sales) have too few comps for a stable median; report n so the caller can weight it.
  • current_bid mid-auction ≠ final price. BaT auctions get most of their action in the final minutes. A live bid far below median with days left is expected, not a deal. Gate "below-market" confidence on days_left (and prefer no-reserve lots).
  • timestamp_end is epoch seconds, and current_bid on completed items is the final hammer price (sold_text distinguishes "Sold for $X" from a no-sale/reserve- not-met listing where sold_text is empty).
  • Autocomplete param is term. ?term= → 200; ?q=/?query=/?keyword= → 400.
  • No anti-bot wall observed. Plain GET (no proxies, no stealth) returned HTTP 200 with the full inline feed across every page tested. --proxies is an optional safety net, not a requirement.

Expected Output

One object; results[] has one entry per target model. Live auctions with below_market: true are the candidates; use days_left + no_reserve to rank confidence.

{
  "success": true,
  "generated": "2026-07-07T23:00:00Z",
  "live_total": 960,
  "results": [
    {
      "target": "Datsun 240Z",
      "model_page": "https://bringatrailer.com/datsun/240z/",
      "market": { "n": 553, "source": "stats.sold(full history)", "p25": 17500, "median": 25000, "p75": 36500 },
      "live_auctions": [
        {
          "title": "1972 Datsun 240Z 4-Speed",
          "current_bid": 22500,
          "ends": "2026-07-08",
          "days_left": 0.8,
          "no_reserve": false,
          "pct_of_median": 90,
          "below_market": true,
          "url": "https://bringatrailer.com/listing/1972-datsun-240z-.../"
        }
      ]
    },
    {
      "target": "Mercedes-Benz W113 SL (Pagoda)",
      "model_page": "https://bringatrailer.com/mercedes-benz/w113/",
      "market": { "n": 555, "source": "stats.sold(full history)", "p25": 49900, "median": 66500, "p75": 88500 },
      "live_auctions": [
        {
          "title": "30-Years-Family-Owned 1970 Mercedes-Benz 280SL 4-Speed",
          "current_bid": 40000,
          "ends": "2026-07-11",
          "days_left": 3.8,
          "no_reserve": true,
          "pct_of_median": 60,
          "below_market": true,
          "url": "https://bringatrailer.com/listing/..."
        }
      ]
    },
    {
      "target": "First-gen Chevrolet Camaro (1967-1969)",
      "model_page": "https://bringatrailer.com/chevrolet/camaro-1967-1969/",
      "market": { "n": 876, "source": "stats.sold(full history)", "p25": 40000, "median": 55000, "p75": 77500 },
      "live_auctions": [
        {
          "title": "1967 Chevrolet Camaro RS/SS Coupe L35 396/325 4-Speed",
          "current_bid": 43150,
          "ends": "2026-07-13",
          "days_left": 5.8,
          "no_reserve": false,
          "pct_of_median": 78,
          "below_market": true,
          "url": "https://bringatrailer.com/listing/..."
        }
      ]
    }
  ],
  "error_reasoning": null
}

Model with no live auctions returns an empty live_auctions: [] (e.g. INEOS Grenadier and Ferrari 365 GTC/4 had zero live lots in the validation run). On total failure: { "success": false, "results": [], "error_reasoning": "<what broke>" }.

Find Below-Market Vintage Cars on Bring a Trailer · browse.sh