walkscore.com

get-score

Installation

Adds this website's skill for your agents

 

Summary

Given a US or Canadian street address (or a walkscore.com /score URL), return the Walk Score, Transit Score, Bike Score, qualitative tier labels, neighborhood/city area label, lat/lon, and canonical URL. Per-category amenity counts available via the browser fallback path.

FIG. 01
FIG. 02
FIG. 03
FIG. 04
FIG. 05
FIG. 06
SKILL.md
258 lines

Walk Score Lookup

Purpose

Given a US/Canadian street address (or any walkscore.com /score/... URL), return the Walk Score (0–100), Transit Score (0–100, when published for the city), Bike Score (0–100, when published), the qualitative tier label for each (e.g. Walker's Paradise, Very Walkable, Somewhat Walkable, Car-Dependent, Rider's Paradise, Biker's Paradise), the area / neighborhood label, geocoded latitude/longitude, and the canonical Walk Score URL. Read-only — never submits feedback, edits a place, or signs in.

Amenity counts: The task prompt lists per-category amenity counts (Restaurants, Coffee, Groceries, etc.). These counts are not rendered in the initial server HTML — they hydrate from a follow-up pp.walk.sc map-marker fetch inside the embedded Google Maps widget. The static HTML only carries the "At this address" featured-place card (a single curated POI). To return per-category counts, fall back to a JS-rendered fetch (browser path below). Default-emit null for unavailable counts and document this in the output.

When to Use

  • A user asks "how walkable is {address}?"
  • Real-estate / rental agents comparing walkability across listings.
  • Pre-filling a relocation worksheet (commute mode mix, transit, bike).
  • Bulk scoring of a list of addresses (the API path costs ~0.8–1.3s/request and tolerates ≥1 req/s).

Workflow

The walkscore.com consumer detail page is fully server-rendered for the score numbers, tier labels, lat/lon, and area label — every value the task asks for is in the initial HTML. Lead with a single HTTP GET to https://www.walkscore.com/score/{slug} via Browserbase's /v1/fetch API (or a plain proxied HTTP client). Do not spin up a CDP browser unless you specifically need amenity-by-category counts (see "Browser fallback" below — they require JS hydration of the embedded Google Maps marker layer).

1. Construct the score URL from the user-supplied address

The site accepts + or %20 for spaces and normalizes free-form addresses via a 301 redirect to the canonical hyphenated slug. You do not need to reproduce the homepage's encodeAddress() function — just URL-encode the user input.

ADDR="1600 Pennsylvania Ave NW, Washington, DC 20500"
SLUG=$(python3 -c 'import urllib.parse,sys; print(urllib.parse.quote_plus(sys.argv[1]))' "$ADDR")
URL="https://www.walkscore.com/score/${SLUG}"

(Canadian addresses follow the same form but must end in …-bc-canada, …-on-canada, etc. The server's 301 handler fills the -canada suffix in for you — see Gotchas.)

2. Fetch the page via Browserbase Fetch API (preferred — no session, no CDP)

curl -fsS -X POST https://api.browserbase.com/v1/fetch \
  -H "x-bb-api-key: $BB_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"url\":\"$URL\",\"proxies\":false}"   # proxies optional; site is lightly bot-walled

Plain proxies:false works for one-off lookups. Set proxies:true only if you see a 403/Akamai-style block (none observed in 4-iter testing — three sequential bare requests all returned 200 in <1.3s).

3. Handle the 301 redirect (Browserbase /v1/fetch does NOT auto-follow)

STATUS=$(jq -r .statusCode <<<"$RESP")
if [ "$STATUS" = "301" ]; then
  CANONICAL=$(jq -r '.headers.Location' <<<"$RESP")
  # Re-fetch CANONICAL with the same call shape
fi

The server applies redirects for: spaces → hyphens, "Avenue" → "ave", "Fifth" → "5th", "Street" → "st", missing zip → nearest canonical with zip, BC-only → BC-Canada, and approximate fuzzy matches (e.g. 4-private-rd-jackson-wy-83001 301s to 125-e-pearl-ave-jackson-wy-83001). Always re-fetch the Location header and use that body for extraction. The redirect chain is at most one hop in observed cases.

4. Extract from the canonical 200 response

All five required values come from straightforward regex / DOM scans on the HTML:

FieldPattern
walk_scoreInteger in <img src="//pp.walk.sc/badge/walk/score/{N}.svg">
transit_scoreInteger in //pp.walk.sc/badge/transit/score/{N}.svg (absent → null, see gotcha)
bike_scoreInteger in //pp.walk.sc/badge/bike/score/{N}.svg (absent → null)
walk_label / transit_label / bike_labelFirst <h5 class='tight-bot'>{LABEL}</h5> appearing after each badge img. Decode &rsquo; → '.
lat / lngdata-lat="{N}" data-lng="{N}" on the .commute-summary element. Stable single occurrence per page.
area_label (neighborhood + city)data-label="{LABEL}" on the same .commute-summary element (e.g. Downtown Washington, DC, Downtown Manhattan, Downtown Vancouver).
address<title>{ADDRESS} - Walk Score</title> — already normalized to the canonical form ("Avenue Northwest" not "Ave NW").
canonical_url<link rel="canonical" href="{PATH}"> (note: site emits a path, not absolute URL — prepend https://www.walkscore.com).

Pseudo-extractor (Node, no DOM lib needed):

const html = response.content;
const num = (type) => parseInt((html.match(new RegExp(`badge/${type}/score/(\\d+)`)) || [])[1]) || null;
const tierFor = (type) => {
  const idx = html.indexOf(`badge/${type}/score/`);
  if (idx < 0) return null;
  const m = html.slice(idx, idx + 2000).match(/<h5\s+class=['"]tight-bot['"]>([^<]+)<\/h5>/);
  return m ? m[1].replace(/&rsquo;/g, "'") : null;
};
const out = {
  walk_score: num('walk'), transit_score: num('transit'), bike_score: num('bike'),
  walk_label: tierFor('walk'), transit_label: tierFor('transit'), bike_label: tierFor('bike'),
  lat: parseFloat((html.match(/data-lat="([^"]+)"/) || [])[1]) || null,
  lng: parseFloat((html.match(/data-lng="([^"]+)"/) || [])[1]) || null,
  area_label: (html.match(/data-label="([^"]+)"\s+class=['"]commute-summary/) || [])[1] || null,
  address: (html.match(/<title>([^<]+)\s*-\s*Walk Score<\/title>/) || [])[1] || null,
  canonical_url: 'https://www.walkscore.com' + ((html.match(/<link[^>]+rel=["']canonical["'][^>]+href=["']([^"']+)/) || [])[1] || ''),
};

5. Branch on the response status

  • 200 + populated badges → success, return scores. If transit_score or bike_score is null, that city simply doesn't have Walk Score's Transit / Bike Score published — return null for both score and label (not zero).
  • 301 → follow Location once, then re-extract.
  • 404 → emit { success: false, error_reasoning: "address_not_found" }. The 404 page title is 404 Page Not Found - Walk Score.
  • 200 + missing badges → only seen when the URL is the lat/lng metro-centroid form (/score/loc/lat=X/lng=Y); treat as success: false, error_reasoning: "address_not_geocoded".

Browser fallback (only when amenity-by-category counts are required)

The amenity counts (Restaurants, Coffee, Groceries, Shopping, Errands, Parks, Schools, Entertainment) come from the Google Maps marker layer that hydrates after initialize() runs. To capture them:

SID=$(curl -sS -X POST https://api.browserbase.com/v1/sessions \
  -H "x-bb-api-key: $BB_API_KEY" -H "Content-Type: application/json" \
  -d "{\"projectId\":\"$BB_PROJECT_ID\",\"keepAlive\":true}" | jq -r .id)
browse env remote
browse --connect "$SID" open "$URL" --wait load
browse --connect "$SID" wait timeout 3000           # markers hydrate ~2–3s after load
browse --connect "$SID" eval '(function(){
  // Sidebar nav exposes category counts as <a class="tab" data-type="X"><span class="count">N</span></a>
  const out = {};
  document.querySelectorAll(".sidebar-nav a.tab, .ws-amenities-tabs li").forEach(el => {
    const t = el.getAttribute("data-type") || el.querySelector(".label")?.textContent?.trim();
    const c = el.querySelector(".count")?.textContent?.trim();
    if (t && c) out[t] = parseInt(c) || 0;
  });
  return out;
})()'

The exact selectors for the amenity tab list can drift; verify against a snapshot first. Do not click amenity tabs / map markers / "Get a quote" buttons — read-only is the rule.

Site-Specific Gotchas

  • /v1/fetch does not follow 301s. The single most common reason a first extraction returns all-nulls is forgetting to follow the redirect — the 301 body is a tiny "Redirecting..." stub with no badges. Always re-fetch on statusCode == 301 and read from the 200 body.
  • Transit / Bike score absence is a real outcome, not an error. Cities without Walk Score's transit data (Jackson WY, most exurbs, many small Canadian cities) simply omit the badge/transit/score/{N} markup. Emit null for both the score and label — do not default to 0.
  • Tier labels use HTML entities. Rider's Paradise ships as Rider&rsquo;s Paradise (likewise Biker&rsquo;s Paradise, Walker&rsquo;s Paradise). Decode &rsquo; → ' before returning. The five Walk Score tiers are Walker's Paradise (90–100), Very Walkable (70–89), Somewhat Walkable (50–69), Car-Dependent (25–49), Car-Dependent again at lower scores ("Almost all errands require a car"). Transit tiers: Rider's Paradise (90+), Excellent Transit (70–89), Good Transit (50–69), Some Transit (25–49), Minimal Transit (<25). Bike tiers: Biker's Paradise (90+), Very Bikeable (70–89), Bikeable (50–69), Somewhat Bikeable (<50). Emit the literal label from the page rather than re-deriving from the integer — the boundaries shift occasionally and the page is authoritative.
  • Canadian addresses need -canada in the slug, but the 301 handler will add it for you. https://www.walkscore.com/score/200-burrard-st-vancouver-bc → 301 → …-vancouver-bc-canada. Don't bother detecting Canadian addresses client-side; just submit the raw form and follow.
  • Server fuzzy-matches addresses very aggressively. 4-private-rd-jackson-wy-83001 (a nonsense rural address) 301s to 125-e-pearl-ave-jackson-wy-83001 (the nearest known address). Returned scores reflect the redirected address, not the original input. Always emit the canonical_url and the extracted <title> address so callers can detect when the lookup snapped to a different point. If address differs meaningfully from the user's input, flag it.
  • /score/loc/lat=X/lng=Y is NOT a per-address lookup. It returns the score for the metro centroid nearest those coordinates, not the building at that point. E.g. lat=40.7580/lng=-73.9855 (real Times Square) returns the generic "Manhattan NY" page with data-lat=40.7208, data-lng=-74.0006 (Tribeca) and metro-default scores 100/100/92. Use this URL form only for "what's it like in {neighborhood}?" queries — never for street-address lookups. Prefer the slug path.
  • <title> city is double-printed. Observed: "1600 Pennsylvania Avenue Northwest, Washington, DC DC - Walk Score" — the state abbreviation appears twice. This is a Walk Score template bug; if you parse city/state from the title, dedupe the trailing token.
  • encodeAddress() is defined inline on the homepage in a onsubmit handler that we did not need to reproduce — URL-encoding the raw user input plus a 301 follow is functionally equivalent and avoids depending on the homepage JS surviving.
  • Lat/lng on the page is the geocoded centroid of the addressed property, not a commute target. The .commute-summary element happens to live in the commute widget but its data-lat/data-lng is the address point. Stable single occurrence per page (one address = one geocode).
  • Amenity counts are not in the static HTML. The static page only carries one "At this address" featured place (curated POI like "White House Rose Garden"). The Restaurants / Coffee / Groceries / Shopping / Errands / Parks / Schools / Entertainment counts come from the Google Maps marker layer (maps.googleapis.com/maps/api/js?...&libraries=places&callback=initialize) populated after a follow-up XHR to walkscore.com — if your task needs them, take the browser-fallback path.
  • No formal rate limit observed. Three sequential bare-IP requests for the same score page returned 200 in 0.7–1.3s each with no captcha. The site's robots.txt and 60-second Cache-Control: public, max-age=60 header suggest sustained ≥1 req/s is fine, but pace politely (the SKILL spec says "lookups should pace politely"). For bulk runs >100 addresses, add a 0.5s jitter and route via proxies:true to avoid surface contention.
  • No bot wall for read-only score pages. GET on /score/... works without --verified or --proxies. Reserve those flags for the browser-fallback path if you hit interactive Akamai.
  • Two distinct &copy; notices in the footer reference Walk Score data and the Google Maps imagery — Walk Score &copy; 2026 is the dynamic year; don't use it as a freshness signal.
  • Do NOT use this skill to submit feedback, edit places, or click the "Get a quote" / "Apartments for rent" leadgen widgets. Those break the read-only contract and trigger downstream affiliate analytics.

Expected Output

{
  "success": true,
  "address": "1600 Pennsylvania Avenue Northwest, Washington, DC 20500",
  "input_address": "1600 Pennsylvania Ave NW, Washington, DC 20500",
  "walk_score": 84,
  "walk_label": "Very Walkable",
  "transit_score": 100,
  "transit_label": "Rider's Paradise",
  "bike_score": 79,
  "bike_label": "Very Bikeable",
  "area_label": "Downtown Washington, DC",
  "latitude": 38.9037406,
  "longitude": -77.0362967,
  "amenities": null,
  "canonical_url": "https://www.walkscore.com/score/1600-pennsylvania-ave-nw-washington-dc-20500",
  "redirected": false,
  "error_reasoning": null
}

Outcome shapes

A. Full-coverage US/CA address (Walk + Transit + Bike all published)

{ "success": true, "walk_score": 100, "walk_label": "Walker's Paradise",
  "transit_score": 100, "transit_label": "Rider's Paradise",
  "bike_score": 92, "bike_label": "Biker's Paradise",
  "area_label": "Downtown Manhattan", "latitude": 40.7208595, "longitude": -74.0006686,
  "canonical_url": "https://www.walkscore.com/score/350-5th-ave-new-york-ny-10118",
  "redirected": true, "amenities": null, "error_reasoning": null }

B. Address with no transit data published (small/exurban city)

{ "success": true, "walk_score": 88, "walk_label": "Very Walkable",
  "transit_score": null, "transit_label": null,
  "bike_score": 74, "bike_label": "Very Bikeable",
  "area_label": "Downtown Jackson", "latitude": 43.4799291, "longitude": -110.7624282,
  "canonical_url": "https://www.walkscore.com/score/125-e-pearl-ave-jackson-wy-83001",
  "redirected": true, "amenities": null, "error_reasoning": null }

C. Canadian address (auto--canada redirect)

{ "success": true, "walk_score": 97, "walk_label": "Walker's Paradise",
  "transit_score": 100, "transit_label": "Rider's Paradise",
  "bike_score": 69, "bike_label": "Bikeable",
  "area_label": "Downtown Vancouver", "latitude": 49.281954, "longitude": -123.1170744,
  "canonical_url": "https://www.walkscore.com/score/200-burrard-st-vancouver-bc-canada",
  "redirected": true, "amenities": null, "error_reasoning": null }

D. Address fuzzy-snapped to a different building (warn caller)

{ "success": true, "input_address": "4 Private Rd, Jackson, WY 83001",
  "address": "125 East Pearl Avenue, Jackson WY",
  "walk_score": 88, "walk_label": "Very Walkable",
  "transit_score": null, "bike_score": 74, "bike_label": "Very Bikeable",
  "canonical_url": "https://www.walkscore.com/score/125-e-pearl-ave-jackson-wy-83001",
  "redirected": true, "warning": "input_address_snapped_to_nearest_known_point",
  "error_reasoning": null }

E. Address not found

{ "success": false, "input_address": "99999 Fakestreet, Fakecity, ZZ 00000",
  "error_reasoning": "address_not_found" }

F. With amenity counts (browser-fallback path only)

{ "success": true, "walk_score": 84, ...,
  "amenities": { "Restaurants": 25, "Coffee": 19, "Groceries": 4,
                 "Shopping": 32, "Errands": 28, "Parks": 17, "Schools": 9, "Entertainment": 21 },
  "canonical_url": "https://www.walkscore.com/score/1600-pennsylvania-ave-nw-washington-dc-20500" }