avito-connector

Avito Connector

Reads Avito via the internal js/items JSON API. Two tiers: impersonated HTTPS (curl_cffi) works from a residential Russian IP; from a datacenter the firewall answers 403 + captcha, so the fetch falls back to the operator's logged-in Chrome over CDP — same pattern as Ozon.

When to use

• Search classifieds by text, with optional category/location filters
• One listing's price, description, views
• Seller reputation — Avito is classifieds: no per-item reviews, the seller's rating and active-listing count ARE the review signal

Tools available

• avito_search(query, page, location_id, category_id, include_inactive) — listings via js/items. price_rub is None for ads with no price (free/exchange) — never 0; during an Avito sale it is the price after the discount, with old_price_rub and discount_percent beside it. Each hit has availability (active/reserved/unknown); reserved hits are kept unless include_inactive=false (then excluded_inactive says how many and _meta carries an info: line).
• avito_card(item_id_or_url, photos=0) — one listing: price, description, views, seller, availability (active/reserved/unknown) with its evidence, from the page's service flags only, never from the seller's title or description. Reserved means another buyer holds it; unknown is not a verdict. Also deal terms (condition, delivery, safe deal, bargain), seller checks and tenure, city and stations (street address only when every source says company), date, views. photos=1..3 attaches listing photos of an active ad (costly in tokens; describe the item, not people, documents or plates in the photos); photos_status says why fewer arrived.
• avito_seller(seller_id_or_url) — seller rating, review count, active listings

Not an MCP tool: avito_selfcheck() is a tri-state drift canary. It is CLI-only — marketplace-mcp doctor runs every connector's canary at once.

What a search row does and does not carry

Confirmed against a real js/items response captured 2026-07-28 from a residential session:

• location is an object upstream, not a string: {"id": 637640, "name": "Москва", ...}. The connector reduces it to the place name. When Avito itself has no place name for an ad, the field is None — the numeric locationId is deliberately NOT resolved to a city, because that would need a lookup table this connector does not have and the result would be a guess that reads like data.
• price does not exist as a top-level key. The amount lives in priceDetailed.value; the connector reads it and leaves None for priceless ads. During an Avito sale that value is the price before the discount; the price a buyer pays is the hit's iva.PriceStep component price, next to a discount component (old_string, percent). A discount that does not add up is price_rub=None with a price_conflict: line in _meta, never a guess.
• posted_at is ISO-8601 UTC, converted from Avito's sortTimeStamp (epoch milliseconds). There is no date string in the payload to pass through.
• seller_name is None on search rows. The payload has no seller object at all — only a logo link. Use avito_card or avito_seller when you need the seller; do not infer a name from the search row.

Gotchas

• A 403 that sets cookies and returns plausible HTML is the firewall, not data; only JSON-shaped 200s count. The connector handles this — trust tier_used.
• location_id defaults to Moscow (637640); pass another Avito location id per call.
• Deep pagination is unreliable, and not because of this connector. Avito escalates 429 then 403 once a session walks past the first page or two; the maintainers of the widely used open-source Avito parsers report the same and advise staying on page 1. Take the first page, narrow the query instead of paging, and treat a rate_limited or transport_down on page > 1 as expected rather than as drift.
• From a datacenter IP expect the CDP tier; keep the scraping-profile Chrome logged into avito.ru.
• The JSON API and the HTML pages are gated differently: js/items can answer from a residential session while avito.ru/<region>/<category>/<slug> still shows the security wall. A blocked listing page does not mean search is down.

DSH activation

In DeepSeek Harness, the default profile exposes only compare_prices and compare_sources through the cheap compare mount. Per-marketplace tools and marketplace_sources require RU_MARKETPLACE_MCP_FULL=1 and a profile restart; do not call them in the default mode.