SuperAPI AI Directory

sutochno

General & other

View repository

Updated 10-15 17:28

About this skill

Sutochno Search

Search Sutochno.ru (СУТОЧНО.РУ) — a Russian short-term-rental marketplace where you book apartments, rooms, and houses directly from owners and pay in rubles, typically via Mir card or bank transfer. It's the natural fit for Russian-speaking travellers booking stays in Georgia / the CIS, where Airbnb and Booking have withdrawn or won't take Russian cards. It complements the Airbnb MCP and the vrbo skill: those reach Western inventory, Sutochno reaches the ruble-priced, owner-direct inventory those platforms miss.

Sutochno is NOT behind Akamai. Unlike the vrbo skill (which fights an Akamai bot wall and must run headed), Sutochno's server-rendered city pages load fine headless — there is no bot sensor to defeat. This skill still ships on the toolkit's standard patchright-docker base (the same image every browser skill builds on), but only for the bundled chromium + OS deps; it does not need Patchright's stealth. Patchright is a drop-in Playwright fork, so the browser API is identical. The default mode is headless; --headed is opt-in for debugging.

The JSON API is auth-walled — we scrape pages instead. Sutochno's /api/* endpoints (e.g. /api/geo/suggest) return 403 Application authentication failed without an app token the SPA computes client-side. So this skill does not touch the API. It scrapes the server-rendered city pages, where the listing cards are already present in the HTML.

City results live on SUBDOMAINS. Each city has its own host: https://tbilisi.sutochno.ru/, https://batumi.sutochno.ru/, https://kutaisi.sutochno.ru/, etc. The country directory https://sutochno.ru/georgia lists the supported Georgian cities and links out to https://{city}.sutochno.ru/<listingId> cards. Pass the city as a subdomain keyword (--city tbilisi) or a full URL (--url).

No login required. Sutochno search results are public; the skill finds rentals and returns links only — it does not log in or complete a booking.

Prerequisites

pip install patchright && patchright install chromium

Or use Docker (no local install):

docker build -t sutochno skills/sutochno/

When to Use

  • A Russian-speaking traveller (or one travelling with Russian companions) booking accommodation that's paid in rubles via Mir card / bank transfer
  • Georgia / CIS stays (Tbilisi, Batumi, Kutaisi, …) where Airbnb / Booking withdrew or won't accept Russian cards
  • Cross-referencing owner-direct ruble inventory against Airbnb / VRBO before booking a multi-night stay

When NOT to Use

  • Completing a booking. This skill finds rentals and returns links only. Do not attempt to complete a purchase or contact the owner.
  • Non-CIS destinations. For Western destinations Airbnb (MCP), the vrbo skill, and Booking have far deeper inventory and card support — use those.
  • Hotels / branded chains. Use compare-hotels / serpapi / liteapi / the portal skills.

Usage

# Tbilisi, human-readable table (headless by default)
python3 skills/sutochno/scripts/search_sutochno.py --city tbilisi

# Batumi, JSON, only 9.0+ rated, under 8000 rubles/night, top 15
python3 skills/sutochno/scripts/search_sutochno.py --city batumi \
  --min-rating 9.0 --max-price 8000 --limit 15 --json

# Explicit full URL (e.g. a city page with query filters already applied)
python3 skills/sutochno/scripts/search_sutochno.py \
  --url "https://kutaisi.sutochno.ru/" --json

# Docker
docker run --rm sutochno --city tbilisi --json

Arguments

Flag Default Notes
--city one of city/url required City subdomain keyword, e.g. tbilisi, batumi. Resolves to https://{city}.sutochno.ru/.
--url one of city/url required Full city URL — use instead of --city to pass a page that already carries query filters.
--min-rating 0 Client-side filter, 0–10 scale (e.g. 9.0). Listings with no rating are kept.
--max-price 0 Client-side per-night ceiling in rubles. Listings with no price are kept.
--limit 20 Max listings returned (after sort).
--headed off Run the browser headed (debugging only — Sutochno has no bot wall, so headless is the default).
--json off Emit JSON to stdout (use this for orchestration).

Selectors (the scrape contract)

On a city page, each listing card exposes:

Field Selector Raw shape Extraction
name + link .card-content__object-type (an <a>) text + href → https://{city}.sutochno.ru/<id> (or /hotels/<id>) name text; id = trailing digits of href
price .card-prices__price "4 890 ₽ за сутки" substring before ₽, strip all non-digits → 4890
rating + reviews .card-prices__rating "10 отзывов 9,8 (10) …" rating = first /\d{1,2},\d/ (comma → dot); reviews = /(\d+)\s*отзыв/
address .card-content__address "Тбилиси, Старый город" as-is

The card wrapper is the nearest ancestor of the name node that also contains a .card-prices element — name and prices are siblings, not nested — so the extractor climbs up from the name link until el.querySelector('.card-prices') is truthy, then reads price/rating from inside that wrapper.

Price gotcha: the thousands separator in the price string is a mixed-width / non-breaking space, so a naive \d+ regex truncates the leading digit (returns 890 for 4 890). The reliable method is to split on ₽ then strip all non-digits — that's what parse_price does.

Images are lazy-loaded placeholders (lzy-img.gif) on first paint, so the scraper does not rely on photos.

Output

JSON shape (with --json):

{
  "source": "sutochno",
  "searchUrl": "https://tbilisi.sutochno.ru/",
  "query": { "city": "tbilisi", "url": null, "min_rating": 9.0, "max_price": 8000 },
  "count": 12,
  "listings": [
    {
      "id": "1234567",
      "name": "Уютная студия в центре",
      "area": "Тбилиси, Старый город",
      "rub_per_night": 4890,
      "rating": 9.8,
      "reviews": 10,
      "url": "https://tbilisi.sutochno.ru/1234567"
    }
  ]
}

Listings are deduped by id and sorted rating desc, then price asc (entries with no rating / price sort last). rub_per_night is the lead nightly rate in rubles — open the listing URL to confirm the all-in total (cleaning, deposit, and any owner fees may not be in the SERP figure).

Tests

The pure logic (city-URL building, ruble price parsing, rating/review parsing, listing-id extraction, normalisation, dedupe / filter / sort) is covered by stdlib unittest tests — no browser, network, or Patchright needed (the patchright import is lazy):

python3 -m unittest discover -s skills/sutochno/tests -v   # or: python3 -m pytest skills/sutochno/tests

The browser-driven path (page navigation, DOM extraction) can't be unit-tested offline — verify it with a real city search (see Usage above).

Known fragility

  • DOM selectors drift. Sutochno may revise the card-content__* / card-prices__* class names. If a live search returns 0 cards on a city you know has inventory, re-inspect a city page and refresh the selectors in EXTRACT_JS and the selector table above (file a P2 task).
  • The API stays auth-walled. Don't be tempted back to /api/* — it needs the SPA's computed app token and returns 403 Application authentication failed without it. The server-rendered pages are the supported surface.