citilink-connector

Citilink Connector

Reads Citilink inside the operator's Chrome over CDP. The whole domain answers HTTP 429 (stricter than DNS), and its data transport is gRPC-web — a binary protobuf protocol that is not worth reversing. A real Chrome passes Qrator and renders the pages, so the connector reads the DOM instead.

When to use

• Search Citilink by text
• One product's price, old price, availability

Tools available

• citilink_search(query) — tiles from the rendered search page
• citilink_card(product_url) — one card. The URL carries /product/<slug>/, where the slug ends in a numeric id: /product/noutbuk-lenovo-2169270/.

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

Why the price used to come back null

Citilink renders the currency glyph in a separate element from the digits:

<span data-meta-price="59990"><span>59 990</span><span>₽</span></span>

The old extractor filtered text lines for руб|₽ and so never saw a line carrying both. Title parsed, price did not — on every product.

Read the data-meta attributes, not the class names

Citilink's class names are build-hashed (app-catalog-51bw0j-…) and change with every deploy, so they are useless as selectors. The site publishes a stable data-meta-* contract for its own analytics, and that is what the extractor keys on:

attribute	holds
data-meta-name="Snippet__title"	the product name anchor
data-meta-name="Snippet__price"	the current price (glyph in a child span)
data-meta-name="Snippet__old-price"	the strikethrough price, no glyph
data-meta-price="59990"	the exact amount, machine-readable

data-meta-price needs no parsing and cannot be confused with an instalment or a badge, so it is preferred; the display strings are the fallback.

If prices start coming back null, check whether those attributes still exist before touching anything else — a class-name change is expected and harmless, a data-meta-* change is the real drift.