# Iskat

> Independent search index of the Russian-language web, built for AI agents.
> 321712 pages from 239 sources.

## Getting access

Agents register themselves — no human step, no email confirmation:

    curl -X POST https://iskat.io/api/agent/register       -H 'content-type: application/json'       -d '{"name":"my-agent","purpose":"what you use it for"}'

The response contains an API key. Limits: 30 requests/minute, 2000/day.

## MCP (preferred)

    https://iskat.io/mcp        Streamable HTTP, header: Authorization: Bearer <key>

Tools:
  search            hybrid keyword+vector search, cross-lingual
  search_products   goods with price, stock, specs and a link to the shop
  search_batch      up to 10 queries in one call
  find_passages     answer-shaped passages with their source
  find_similar      pages close to a given one
  fetch_page        full cleaned text from the index, no visit to the origin
  fetch_url         live fetch of a page outside the index (private hosts blocked)
  source_profile    what a source is: type, authority, coverage, freshness
  watch_query / check_watch / list_watches   ask what is new on a topic
  submit_site / submission_status            propose a source, track it
  list_sites / index_stats                   what is already indexed

## REST

    GET /api/search?q=...&mode=hybrid&category=...&limit=10
    GET /api/products?q=...&max_price=...&in_stock=true&sort=price_asc
    GET /api/page?url=...        full stored text, no request to the origin
    GET /api/similar?url=...
    GET /api/categories

## What makes this different from a normal search API

fetch_page returns the complete cleaned article text from our index. The agent
does not visit the original site, so blocked crawlers, paywalls, JS rendering
and slow pages stop being its problem.

Retrieval is cross-lingual: a Russian query matches English documents and the
other way round.

## Growing the index

Coverage is curated, not exhaustive. If a search finds nothing on a topic you
need, call submit_site with the source that would have answered it. Submissions
are checked against robots.txt and reviewed before crawling.