Making 1,200 static pages readable by AI agents: MCP tools + NLWeb on Cloudflare Pages Functions
I run OneFindMe , a free AliExpress product search that works in 21 languages. Over the past months the site grew a library of about 1,200 static pages: size guides, import-tax pages per country, buying guides, home-problem fixes, all written per language. The site already had an MCP server at https://onefindme.com/mcp with a search_products tool. It is listed in the official MCP registry and in…
OneFindMe, a free AliExpress product search site, grew a library of about 1,200 static pages in 21 languages. The site already had an MCP server with a search_products tool, but there was a gap: an assistant connected to it could search products, but could not read any of the guides. To close this gap without using a database, a vector store, or a second service, the following steps were taken:
1. A metadata index was built at deploy time. A small Python script walked sitemap.xml, skipped noindex pages and search home pages, and kept only the information each page already published about itself. This included the page URL, language, title, meta description, headings, and FAQ questions. The generated index consisted of 1,213 entries, about 900 KB, written to functions/_guides.js as export default [...]. The full text of the pages was not stored.
2. Two tools were created: search_guides and get_guide. The search function used keyword scoring with field weights to rank the results. The titles, headings, and FAQ questions were written in the reader's language, so they contained the words people typed. The get_guide function took a URL from the search results and returned the page as Markdown. The index held no text, as the function read the live page through the static-asset binding and converted it on the fly.
3. An NLWeb endpoint was created using Microsoft's open protocol for asking a website a question in natural language. The function returned schema.org-shaped results, with information about the URL, title, site, score, and description. The answer could never be stale than the page itself, as it was always fetched on demand.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.