HappyCow Vegan Search

Purpose

Search HappyCow for vegan, vegetarian, and veg-friendly restaurants — plus health-food stores, vegan stores, and other veg-friendly venues (bakeries, juice bars, ice cream, coffee) — in any location, honoring the full filter surface (diet/venue type, cuisine, features, price, rating, distance, sort) and return matches as structured JSON. Read-only — never clicks Add a Place, Write a Review, Sign In, Bookmark, or any mutation control.

When to Use

• "List vegan and vegetarian restaurants within 10 km of Berlin, sorted by rating."
• "Find Italian-cuisine vegan spots with outdoor seating in Brooklyn."
• "Map all health-food stores within 5 miles of Austin, TX."
• Any read-only HappyCow listing extraction — pulling the full venue catalog for a metro, neighborhood, or lat/lng radius.

Workflow

HappyCow's /searchmap?... UI hydrates the venue list via an internal XHR (GET /ajax/views/searchmap/venues?...). That XHR is confirmed-blocked from cookieless fetch (returns the HappyCow 404 page; requires PHPSESSID + X-Requested-With: XMLHttpRequest from a hydrated browser session — see Site-Specific Gotchas). The good news: HappyCow ships a server-rendered printable view at /searchmap/print?... that honors every filter the UI exposes and returns a static HTML list of all matching venues (paginated ~180/page). Lead with that; fall back to scripted browsing only when the print path is unreachable.

Anti-bot: HappyCow is behind Imperva (Incapsula). A bare browse cloud fetch is 403'd at the bot-check; browse cloud fetch --proxies (Browserbase residential proxy) returns 200. For browser sessions, use both --verified and --proxies.

Step 1 — Build the query

Set up a Browserbase session config (proxies mandatory, Verified recommended) for any browser fallback. For pure fetch, no session is needed — just browse cloud fetch --proxies.

# Browser fallback session (only if print path fails)
SID=$(browse cloud sessions create --keep-alive --verified --proxies | jq -r '.id')

Map the user's location + filters → URL params. The full param schema (extracted from /js/modules/happycow.common.search.map.js + the search form):

Diet / venue-type filter IDs (filters= is dash-joined; each token corresponds to a button name= in the form):

Restaurants: vegan, vegetarian, vegfriendly, veganprofessional (chains), delivery, catering, foodtruck, chains. Stores & more: bakery, coffee, juicebar, icecream, health (health store), vegshop, veganStores (vegan-only stores), farmers, marketvendor, bnb, spa, organization, other.

Cuisine / food-type IDs (ft= is comma-joined; values verified against the in-page filter buttons):

Feature / amenity IDs (f= is comma-joined):

Rating / Open-now / Price — the searchmap UI does not expose rating-min, "open now" boolean, or a price filter as standalone URL params. Sort by order=rating and filter the response. Open-now is encoded through the time-slider as openAt=<HHMM>&openAtDay=<Mon|Tue|...> (separate handling from a simple open=1 flag). Price tier comes back per-venue from the detail page (itemprop="priceRange") — filter client-side.

Step 2 — Fetch the printable list (the optimal path)

QS="location=Berlin%2C+Germany&metric=km&radius=10&filters=vegan-vegetarian&order=rating"
browse cloud fetch --proxies \
  "https://www.happycow.net/searchmap/print?${QS}&page=1" \
  --output /tmp/hc-print-p1.html

The HTML is server-rendered. Key extractable fields per venue card:

• Name + diet type — heading line <NAME> (<DIET>) <distance> <unit> where <DIET> is one of Vegan, Vegetarian, Veg-options, Health Store, Veg Store, Vegan Store, Other. HTML entities (' for ', & for &) appear in names — un-escape.

• Distance — number + unit (km or mi) immediately after the diet paren.

• Rating (stars) — <ul class="rating-list"> with exactly five <li> children. Each <li> contains one <path d="...">. The first ~30 chars of the d attribute identify the star state:

  • M480-219.913 → full star (+1.0)
  • M358.544-253.457 → half star (+0.5)
  • M480-627.109 → empty star (+0)

  Sum across the 5 li → rating (0.0 – 5.0).

• Address + phone — <p class="mt-3 font-normal">Street, City, Country, Postal - Telephone: +CC-...</p>. When phone is unlisted the line reads ... - Telephone: N/A. Card boundary: split the HTML on <div class="mt-3 flex items-center">.

The page header carries two crucial scope fields:

• Heading — <h3>Vegetarian Restaurants & Health Food Stores</h3> (or store-only/restaurant-only variant depending on filters).
• Radius line — <p>Within a radius of <X> <km|mi> of your location - Total Results: <N></p>. This gives you total_results for the region.

Step 3 — Paginate

The print page returns ~180 cards/page (180 was observed for Berlin page 1 and page 2 with radius=10&metric=km, with Total Results: 1833). Iterate page=1, page=2, ... until you've consumed Total Results / 180 pages (round up), or until a page returns zero cards.

for PAGE in $(seq 1 N); do
  browse cloud fetch --proxies "https://www.happycow.net/searchmap/print?${QS}&page=${PAGE}" \
    --output "/tmp/hc-print-p${PAGE}.html"
done

Step 4 — Enrich (optional, only when caller needs venue_id, slug, lat/lng, photos, hours, website, features, cuisine tags, claimed-by-owner)

The print view does not include venue IDs, slugs, lat/lng, photos, or feature/cuisine tags — only what's listed above. For richer fields, two enrichment paths:

(a) City-slug landing pages — /europe/<country>/<city>/, /north-america/usa/<state>/<city>/, /asia/<country>/<city>/. These pages render ~50 curated venue cards as <a href="/reviews/<slug>-<venue_id>">...</a> anchors with the data-* map markers (data-lat, data-lng, data-vegonly, data-new) and full thumbnail URLs. Best when the caller's location is a known city and they want top venues with full IDs.

browse cloud fetch --proxies "https://www.happycow.net/europe/germany/berlin/" \
  --output /tmp/hc-berlin-slug.html
# Extract: grep -oE 'href="/reviews/[a-z0-9-]+-(\d+)"' yields slug→venue_id pairs (~54 unique for Berlin).

Then build the canonical URL: https://www.happycow.net/reviews/{slug}-{venue_id}.

(b) Venue detail page — /reviews/{slug}-{venue_id} returns full schema.org Restaurant microdata plus HappyCow data-* attrs. Fields available (verified on /reviews/19-77-vegan-diner-bar-ramones-museum-berlin-414540):

Detail-page fetches are slow (~880 KB HTML each) and rate-cost adds up — only enrich when the caller asks for fields beyond the print summary.

Step 5 — Emit JSON

Return the schema in Expected Output below. Populate bounding_box from the print page's "X km of your location" header + the geocoded center if you can recover lat/lng (the /searchmap?... HTML embeds data-lat/data-lng on #location after geocode — browse cloud fetch --proxies the /searchmap?location=... URL once to grab those, then build bbox from radius).

Step 6 — Release session (browser fallback only)

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

Browser fallback (only if /searchmap/print returns 403 / 5xx)

If the print path is unreachable (Imperva rate-limit, server outage), load the live /searchmap?... URL in a Verified+proxy browser session and read the rendered venue list. The XHR endpoint /ajax/views/searchmap/venues is confirmed-blocked to cookieless GETs (see gotcha), so this path requires actually rendering the page with JS.

SID=$(browse cloud sessions create --keep-alive --verified --proxies | jq -r .id)
browse cloud browse --connect "$SID" open "https://www.happycow.net/searchmap?${QS}"
browse cloud browse --connect "$SID" wait load
browse cloud browse --connect "$SID" wait timeout 4000   # XHR hydration lands 2–4s after load
browse cloud browse --connect "$SID" snapshot            # venue cards now in the a11y tree

Then extract from data-id/data-lat/data-lng on rendered cards. The XHR response is also visible in browser-trace CDP Network.responseReceived events under /ajax/views/searchmap/venues. Per-card parsing of the rendered list mirrors the venue-detail microdata schema above.

Site-Specific Gotchas

• Imperva (Incapsula) anti-bot is the universal gate. Bare browse cloud fetch of /searchmap?... returns a 403 with an _Incapsula_Resource iframe. browse cloud fetch --proxies (residential proxy) returns 200 for /searchmap?..., /searchmap/print?..., /europe/.../<city>/, and /reviews/{slug}-{id} — proxies are mandatory for every fetch, not just session-mode. Verified 2026-05-16 on Berlin and Austin queries.
• /ajax/views/searchmap/venues is confirmed-blocked for cookieless XHR. Reproduced direct POST to the Browserbase Fetch API with X-Requested-With: XMLHttpRequest + Accept: application/json + correct Referer + User-Agent — every variant returned the HappyCow 404 "Page Not Found" HTML. Endpoint requires a PHPSESSID cookie established by a prior /searchmap?... page load (and likely a CSRF/JS-derived token verified by the Laravel middleware). Don't waste turns hand-crafting cookieless XHRs to this endpoint — it doesn't work. The optimal path is /searchmap/print?...; the browser fallback exercises the XHR through a real session, where cookies + headers are set automatically.
• The print page does NOT include venue IDs, slugs, lat/lng, photos, or tag chips. Only name, diet type, distance, 5-star rating, full address, and phone. If the caller needs richer fields, enrich via city-slug pages (for top ~50 venues with IDs) or per-venue detail pages.
• Default radius is 15 (units), not from limit_distance. Omitting both radius and a bbox returns the server's default 15-of-the-selected-metric. Use radius=N&metric=km|mi explicitly. The deprecated limit_distance= parameter from older URLs is silently ignored by the print route.
• The form's s= URL param is a view-state byte, not a diet filter. The page sets s=2 for the user-visible search panel; the internal AJAX rewrites it to s=3. Diet filtering is via filters=vegan-vegetarian-... (dash-joined), not via s. A request without filters= returns all venue types.
• HappyCow geolocates a location= string server-side and embeds the resolved data-lat/data-lng on the searchmap page's #location input. To recover the geocoded center, scrape that input. Free-form like "Shibuya, Tokyo" and "Austin, TX" and "78704" all resolve cleanly. Direct lat/lng pairs work too — pass lat=&lng=&radius=&metric= and skip location=.
• The bb (bounding-box) param has signature swLat,swLng,neLat,neLng (south-west then north-east, per /js/search.map.js locationEl.setAttribute('data-bounds', ...)). When bb is present, lat/lng/radius are still useful for distance-from-center sort but the bbox is the authoritative spatial scope. Required when the user provides a searchmap URL with bb=.
• Pagination on /searchmap/print returns ~180 cards/page regardless of the limit UI hint. limit=18|27|81 is the searchmap map UI's visible-marker cap, not the print page's serving size. Iterate page=1, page=2, ... until you've consumed Total Results / 180 pages.
• HappyCow rating extraction is path-based, not color-based. All five <li> star SVGs use class="text-yellow-500" regardless of fill state. Distinguish state by the first 12 chars of the <path d="..."> attribute: M480-219.913 = full, M358.544-253.457 = half, M480-627.109 = empty. Sum full + 0.5 × half for the rating.
• Diet type spelling in print HTML uses six values, not the form-name set. The card heading parens contain one of: Vegan, Vegetarian, Veg-options (≈ veg-friendly), Health Store, Veg Store, Vegan Store, Other. Map these to your output diet field; do not expect Veg-friendly literally (it's spelled Veg-options in the print view).
• HappyCow domains the world differently from the request IP. A US-proxy browse cloud fetch of /searchmap?location=Berlin%2C+Germany correctly returns Berlin venues — the geocode is location-driven, not IP-driven. No need to swap proxy regions per locale. The page does include a data-is-user-in-europe flag for GDPR but it doesn't affect search results.
• api.happycow.net is NOT the venue API. It's a Symfony API Platform serving the shop products, user auth, top-ambassadors, venue claims, and a few partner-pricing resources. The venue-search resources are not exposed there. (Verified by hitting https://api.happycow.net/index.json — 36 resource classes listed, none of them venue-search.)
• map.happycow.net is a tile / cluster server, not a JSON places API. Direct GET returns 500 with no usable schema for skill purposes.
• Free-form keywords use term=, comma-separated, not space-joined. The UI's "Add keyword" widget appends each term as a separate button; the URL joins them with commas: term=ramen,gluten-free. Single-word term=ramen works too.
• "Open now" is not a single boolean flag. The UI exposes a time-slider that emits openAt=<HHMM>&openAtDay=<Mon|Tue|...> URL params. For "open right now" use the current local time at the search location and current day-of-week.
• Read-only. Do not click Add a Place, Write a Review, Bookmark, Sign In, or any anchor under /reviews/.../write or /members/.... The /reviews/{slug}-{id}/write URL is a review-submission form, not a read-only detail view.
• Region-wide totals are reliable; "matched cards count" can lag. The header's Total Results: N is the authoritative region-wide hit count. The visible cards on a single page are clamped at ~180, so a page-1 fetch alone is not the full picture for any metro with > 180 hits.
• Sandbox limitation observed during build: Browserbase regional connect endpoints (connect.usw2.browserbase.com etc.) did not resolve via DNS from this build sandbox, so the browser-fallback step was validated only by code inspection of the JS bundles, not by live execution. The browse cloud fetch --proxies path was fully validated end-to-end. If you run from a less-restricted environment, the browser fallback should work — but /searchmap/print is the recommended path regardless.

Expected Output

{
  "success": true,
  "search": {
    "location_query": "Berlin, Germany",
    "resolved_center": { "lat": 52.52, "lng": 13.405 },
    "bounding_box": { "sw_lat": 52.43, "sw_lng": 13.27, "ne_lat": 52.61, "ne_lng": 13.54 },
    "radius": 10,
    "metric": "km",
    "filters": {
      "diet": ["vegan", "vegetarian"],
      "cuisines": [],
      "features": [],
      "rating_min": null,
      "open_now": false,
      "price_max_tier": null
    },
    "sort": "rating",
    "page": 1
  },
  "total_results": 1833,
  "page_size": 180,
  "venues": [
    {
      "venue_id": 19181,
      "slug": "cafe-vux-berlin-19181",
      "name": "Cafe Vux",
      "diet": "Vegan",
      "diet_raw": "Vegan",
      "cuisine_tags": ["Cafe", "Bakery", "International"],
      "feature_tags": ["Outdoor seating", "Take-out", "Free Wi-Fi"],
      "rating": 4.5,
      "review_count": 142,
      "price_range": "Inexpensive",
      "address": {
        "street": "Wipperstr 14 (at Karl-Marx-Strasse)",
        "locality": "Berlin",
        "region": null,
        "postal_code": "12055",
        "country": "Germany",
        "full": "Wipperstr 14 (at Karl-Marx-Strasse), Berlin, Germany, 12055"
      },
      "phone": "+49-30680730555",
      "website": "https://cafevux.de",
      "lat": 52.4747,
      "lng": 13.4392,
      "distance": 4.21,
      "distance_unit": "km",
      "hours_summary": "Open Tue-Sun 09:00-20:00. Closed Mon.",
      "primary_photo_url": "https://images.happycow.net/venues/500/19/18/hcmp19181_1234567.jpeg",
      "additional_photos": [],
      "claimed_by_owner": false,
      "recently_added": false,
      "url": "https://www.happycow.net/reviews/cafe-vux-berlin-19181",
      "_source": "print+detail"
    },
    {
      "venue_id": null,
      "slug": null,
      "name": "Cookies Cream",
      "diet": "Vegetarian",
      "diet_raw": "Vegetarian",
      "cuisine_tags": null,
      "feature_tags": null,
      "rating": 5.0,
      "review_count": null,
      "price_range": null,
      "address": {
        "street": "Behrenstr 55 (at above Crackers)",
        "locality": "Berlin",
        "region": null,
        "postal_code": "10117",
        "country": "Germany",
        "full": "Behrenstr 55 (at above Crackers), Berlin, Germany, 10117"
      },
      "phone": "+49-30680730448",
      "website": null,
      "lat": null,
      "lng": null,
      "distance": 1.26,
      "distance_unit": "km",
      "hours_summary": null,
      "primary_photo_url": null,
      "additional_photos": [],
      "claimed_by_owner": null,
      "recently_added": null,
      "url": null,
      "_source": "print"
    }
  ]
}

Notes on the schema:

• _source indicates which path supplied this row. "print" rows have everything the print page yields and null for the enrichable fields; "print+detail" rows were merged with a /reviews/{slug}-{id} detail fetch; "city-slug" rows came from /europe/.../<city>/ and have IDs but limited per-venue depth.
• diet is normalized (vegan / vegetarian / veg-friendly / health-store / veg-store / vegan-store / other); diet_raw preserves the source spelling (Veg-options, etc.).
• phone is null when print shows "Telephone: N/A".
• When bb= was supplied or computed, bounding_box is populated; when only radius/metric were used, compute it from resolved_center ± radius (rough km↔degree conversion).
• For "candidate" / unverified status: if the print path 403s and the browser fallback can't be reached (e.g., DNS-restricted sandbox), emit {"success": false, "reason": "anti_bot_wall", "diagnostic": "..."} rather than partial data.