MCP Page Finder Extractor

What mcp-page-finder-extractor does

The mcp-page-finder-extractor MCP server connects an MCP client to the Mamba Labs Page Finder and Extractor actor on Apify. It accepts one domain, several domains, or company records containing fields such as name, country, ISIN, ticker, or URL. The find_company_page tool then searches the company’s own public website for selected page types.

The available page types include pricing, free trials, demo requests, investor relations, annual reports, security and compliance pages, privacy and legal pages, careers and job boards, documentation, API references, integrations, blogs, press rooms, case studies, events, marketplaces, support centers, login pages, sustainability, diversity, and more. There are 46 types in total. Discovery vocabulary covers 11 European languages.

How it works

Use mode: "locate" to receive page URLs and discovery metadata, or mode: "locate_and_extract" to read the located pages and produce structured fields. Results contain one flat row per input. For each requested type, the output can include the URL, whether the page was found, the discovery method, and a confidence score associated with that method.

A false result means the site was read and its links, sitemap, and related paths did not reveal the page. A null result means the server could not read enough to determine the answer. Check coverage and fetch_status before treating a negative result as conclusive. Extracted fields are returned in a separate findings dataset keyed by the input.

Setup and configuration

Run the package with:

npx -y @mambalabsdev/mcp-page-finder-extractor

Set APIFY_TOKEN to an Apify API token. The server starts an Apify actor run and returns its dataset. The mcp-page-finder-extractor MCP server can be configured as a standard stdio server; the README provides a Claude Desktop configuration using npx, the package name, and the token in the client environment.

Optional inputs include pageTypes, knownUrls, extraction field selections, language hints, concurrency, request ceilings, candidate-page limits, browser rendering, and cache control. knownUrls can bypass discovery for page types whose URLs are already available. Setting skipCache to "true" requests a fresh crawl instead of using the 14-day cache.

Tools and capabilities

• find_company_page: locate or extract requested page types for domains, domain lists, or company identities.
• Return the method used, including known URL, homepage or footer anchor, section hop, sitemap, or path guess.
• Use browser rendering for publicly served pages that require JavaScript when allowRender permits it.
• Process multiple companies concurrently while retaining one request at a time per company.
• Supply page-type-specific and page-agnostic extraction fields.

Limitations and notes

The mcp-page-finder-extractor MCP server is read-only. It reads publicly available pages on the company’s own website and writes nothing elsewhere. It follows each host’s robots.txt rules and crawl delay, identifies itself with a descriptive user agent, and does not sign headers, impersonate a browser, or retry around access blocks. Browser rendering is not used against access controls.

Usage is billed through Apify credits. Locate, extraction, and company-name identity resolution have separate event charges; identity resolution is only used with the companies input. Completed looks are billed even when no page is found, while dead domains, refusals, and robots exclusions return found: null without a locate charge.

The contact and about page types may return personal data that the company publishes on those pages. Such records include is_personal_data and a lawful_basis. The tool does not infer attributes or perform outside lookups, and it is not intended as a general employee-roster extractor.