MCP server

New launch

MCP server

An MCP server that lets an assistant query the board directly instead of scraping it. Live, read-only, no API key.

Connect

Streamable HTTP transport, no authentication:

https://rankcert.com/api/mcp

In Claude Code: claude mcp add --transport http rankcert https://rankcert.com/api/mcp. In any other client, add it as a remote HTTP MCP server with that URL.

Tools

  • board_this_week - the board for a launch week, each product with its verification tier and rank score
  • product_signals - one product with every signal behind its tier: kind, band, source and the date it was last confirmed
  • search_products - search live products by free text, tag or verification tier
  • verification_stats - the tier split and proof-source breakdown across the whole board - the data behind /stats

Also available

  • Every page has a markdown twin under /md/ that an assistant can fetch directly.
  • /llms.txt indexes the whole site in one request.
  • Product pages carry Product and Dataset JSON-LD with the signals, sources and dates.

Design constraints

  • Read-only. Nothing in the MCP surface writes to the board.
  • Never exposes an exact revenue figure or anything identifying a maker's customers - the band is the finest granularity that exists anywhere.
  • Every response carries the verification date, so an assistant can say how fresh the claim is.

Want a specific tool added? hello@rankcert.com.

Why an MCP server rather than a scraper

Assistants already read this site. They arrive through search, parse the HTML, and answer questions about which products are verified and what the evidence says. That works and it works badly: the parse is lossy, the structure that carries the meaning is discarded, and there is no way for the model to ask a follow-up question without fetching another page and guessing at its shape.

An MCP server fixes the shape problem. The board, a product's signals with their sources and dates, search across live products by tag and category, and the aggregate verification statistics are all exposed as typed tools an assistant can call directly. The model asks a question and gets structured data back, rather than reconstructing it from a page that was designed for a person.

It matters more here than on most sites because the interesting part of this data is provenance. Every signal carries a source and a verification date, and those are exactly the fields that get lost when a page is flattened into text. A model that can read them can tell a user that a revenue figure was confirmed from a processor connection last Tuesday, which is a meaningfully different answer from repeating the number.

The same data is already available in less convenient forms - JSON-LD on every product page, markdown twins under /md, an RSS feed and a machine-readable index at /llms.txt - and those are not going away. The MCP server is for the case where an assistant needs to ask several questions in sequence rather than read one document.

Common questions

Do I need an API key?
No. The exposed data is the same public information already on the site, so the server is readable without authentication. Point any MCP client at https://rankcert.com/api/mcp over streamable HTTP.
Can I use it to submit a product?
Not through MCP. Submission and verification run through the authenticated CLI and web flows.