tesla.com logo
tesla.com

find-lowest-used-car-price

Installation

Adds this website's skill for your agents

 

Summary

Given a ZIP code, search distance in miles, and a model filter (m3/my/ms/mx), return the lowest-priced used Tesla within range — price, year, trim, odometer, location, VIN, and listing URL.

FIG. 01
FIG. 02
FIG. 03
SKILL.md
189 lines

Find Lowest-Priced Used Tesla Near a ZIP

Purpose

Given a ZIP code, a search distance in miles, and a used-car filter (the Tesla model — m3, my, ms, mx), return the lowest-priced used Tesla within that radius: its price, year, trim, odometer, location, VIN, a link to the listing, and the total number of matching cars in range. Read-only — never configures an order, reserves, or purchases.

The Tesla used-inventory web UI is a thin client over a public JSON API (/inventory/api/v4/inventory-results). The recommended method calls that API directly from inside a stealth browser session (so Akamai's _abck cookie is present) — one HTTP round-trip returns the fully-sorted result set. A pure browser-DOM fallback is documented at the end.

When to Use

  • "What's the cheapest used Model 3 within 200 miles of 90210?"
  • Monitoring the floor price of used Tesla inventory for a metro area over time.
  • Comparing the lowest available price across models (m3 vs my vs ms vs mx) for one location.
  • Any flow that needs the cheapest in-range used Tesla without going through the order/reserve UI.

Workflow

Anti-bot reality: tesla.com sits behind Akamai. A cookieless request to the inventory API returns HTTP 429 (Server: AkamaiGHost), and the homepage probe returns 403. You must first load a Tesla page in a stealth Browserbase session (--verified --proxies) to obtain the _abck cookie, then issue the API call from page context so that cookie is attached. From a warmed session the API returns HTTP 200.

Geo-scoping reality (critical): the API scopes results by lat/lng, not by the bare zip. Sending only zip+range is silently ignored — results fall back to the session-IP location (with residential proxies this is wherever the proxy egresses, e.g. an Oregon ZIP), and range is treated as 0. You must geocode the ZIP to lat/lng/state yourself and pass them in the query. Once lat/lng/region are present, range filters correctly (verified: San Diego, ~120 mi from 90210, appears at range:200 but not at range:100).

1. Open a stealth session and warm Akamai

sid=$(browse cloud sessions create --keep-alive --verified --proxies \
  | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
browse open "https://www.tesla.com/inventory/used/m3?arrangeby=plh" --remote --session "$sid"
browse wait load --remote --session "$sid"

arrangeby=plh = "price low → high". Use the target model in the path so the warmed page matches the query you'll run.

2. Geocode the ZIP → lat / lng / state

# From page context (avoids sandbox egress restrictions):
browse eval --remote --session "$sid" '(async()=>{
  const g = await (await fetch("https://api.zippopotam.us/us/90210")).json();
  const p = g.places[0];
  return JSON.stringify({lat:+p.latitude, lng:+p.longitude, region:p["state abbreviation"]});
})()'
# → {"lat":34.0901,"lng":-118.4065,"region":"CA"}

api.zippopotam.us is a no-key US ZIP geocoder. Any equivalent geocoder works — you only need decimal lat, lng, and the 2-letter state abbreviation.

3. Call the inventory API from page context

Build the query JSON, URL-encode it, and fetch it inside the page (so the _abck cookie rides along):

browse eval --remote --session "$sid" '(async()=>{
  const zip="90210", range=200, model="m3";
  const lat=34.0901, lng=-118.4065, region="CA";       // from step 2
  const q = {query:{model, condition:"used", options:{},
    arrangeby:"Price", order:"asc", market:"US", language:"en",
    "super_region":"north america", zip, range, lat, lng, region},
    offset:0, count:1, outsideOffset:0, outsideSearch:false};
  const url = "https://www.tesla.com/inventory/api/v4/inventory-results?query="
    + encodeURIComponent(JSON.stringify(q));
  const j = await (await fetch(url,{headers:{accept:"application/json"}})).json();
  const c = (j.results||[])[0] || null;
  return JSON.stringify({total:j.total_matches_found, cheapest:c});
})()'

Query-field meanings:

  • model — the used-car filter: m3 (Model 3), my (Model Y), ms (Model S), mx (Model X).
  • condition:"used" — used inventory only (not new).
  • arrangeby:"Price", order:"asc" — cheapest first, so results[0] is the lowest price.
  • zip + range — ZIP is cosmetic/echo; range (miles) is the search radius and only filters when lat/lng are also present.
  • lat/lng/region — the real geo-scope (from step 2).
  • outsideSearch:false — stay within range. Set true and it ignores the radius and returns cheapest cars nationwide (in-range ones aren't guaranteed first).
  • count — page size. count:1 is enough for the single cheapest car; raise it to list more.

4. Read the result

  • total_matches_found — number of used cars of that model within range.
  • results[0] — the lowest-priced car. Useful fields: Price / TotalPrice / InventoryPrice (all the listed price in USD), Year, TrimName, Odometer + OdometerType, City, StateProvince, VIN, Model.
  • Listing URL: https://www.tesla.com/{model}/order/{VIN}?titleStatus=used.
  • If total_matches_found is 0 / results is empty → no used cars of that model in range (valid result, not an error).

5. Release the session

browse cloud sessions update "$sid" --status REQUEST_RELEASE

Browser fallback (no API)

If the API path is unavailable, the warmed inventory page itself renders the sorted results. With arrangeby=plh the first result card is the cheapest. Because lat/lng can't be set from the URL (the page geocodes the ZIP internally and resets an unrecognized URL zip to the session-IP location), you must set the ZIP through the on-page location control (the header button showing the current ZIP), let the grid re-render, then read the first card's price/VIN with browse get text body / browse snapshot. This is slower and needs the location UI to cooperate — prefer the API path.

Site-Specific Gotchas

  • Cookieless API = 429. A direct browse cloud fetch/curl of /inventory/api/v4/inventory-results returns HTTP 429 from AkamaiGHost (and sets a decoy _abck). You must warm a real stealth session and call from page context. Stealth --verified --proxies was required for the working run.
  • lat/lng drive geo-scoping, not zip. Sending only zip+range is silently ignored: results default to the session-IP location and range collapses to 0. Two different ZIPs with no lat/lng returned byte-identical result sets in testing. Always geocode and pass lat/lng/region.
  • Residential proxy skews the IP fallback. With --proxies on, the "default" location is the proxy's egress ZIP (observed 97818, Oregon), not your real location — another reason to always pass explicit lat/lng.
  • URL zip/range params don't stick. Navigating to .../used/m3?zip=90210&range=200 gets rewritten to ?zip=<proxy-ip-zip>&range=0 after load. The URL is not a reliable way to set location; use the API query fields (or the on-page location control for the fallback).
  • outsideSearch:true ignores the radius. It returns the cheapest cars nationwide and does not guarantee in-range cars sort first (e.g. a Baltimore car appeared above San Diego for a 90210 search). Use outsideSearch:false for a true "within X miles" answer.
  • Top-level count key is total_matches_found (not total_matches / total).
  • Price fields. Price, TotalPrice, and InventoryPrice were all equal in observed used listings (the listed sale price, before fees/taxes/destination). PurchasePrice may add fees. Use TotalPrice for the headline number.
  • Model codes: m3=Model 3, my=Model Y, ms=Model S, mx=Model X. Path segment and query model use the same code.
  • Empty result is legitimate. Rare/expensive models in a small radius (e.g. used Model S within 50 mi of 94043) return total_matches_found:0. Report lowest_price: null, not an error.
  • API version. v4 is current (/inventory/api/v4/inventory-results). Older docs reference v1/v3; if v4 404s, the query-JSON shape is otherwise compatible across versions.

Expected Output

// Matches found — the lowest-priced used car in range
{
  "success": true,
  "zip": "90210",
  "distance_miles": 200,
  "model": "m3",
  "total_matches": 63,
  "lowest_price": 23500,
  "currency": "USD",
  "vehicle": {
    "year": 2020,
    "trim": "Long Range All-Wheel Drive",
    "odometer": 95548,
    "odometer_unit": "Miles",
    "city": "San Diego",
    "state": "CA",
    "vin": "5YJ3E1EB8LF647704"
  },
  "listing_url": "https://www.tesla.com/m3/order/5YJ3E1EB8LF647704?titleStatus=used",
  "error_reasoning": null
}

// No used cars of that model within range (valid, not an error)
{
  "success": true,
  "zip": "94043",
  "distance_miles": 50,
  "model": "ms",
  "total_matches": 0,
  "lowest_price": null,
  "vehicle": null,
  "listing_url": null,
  "error_reasoning": null
}

// Blocked / failed (e.g. Akamai 429 from a non-warmed session)
{
  "success": false,
  "zip": "90210",
  "distance_miles": 200,
  "model": "m3",
  "error_reasoning": "Inventory API returned 429 (AkamaiGHost) — session was not warmed with a Tesla page load / stealth flags."
}