Product Page Extractor: remote MCP server

Give it a product page URL, or HTML you already have, and get back clean JSON: name, brand, price, currency, stock, SKU/GTIN/MPN, images, rating, and every offer.

What makes it different

• Shows where the price came from. Each result has a price_source field (json-ld, microdata, meta, or null), so your agent can tell a structured price apart from a guess. If a page renders its price with JavaScript, you get null rather than a wrong number.

• Works on raw HTML, not just URLs. Already scraped the page, or the site blocks bots? Pass the html (up to 2 MB) and skip the fetch.

• Returns every offer, not just the first price: variants, sellers and sale prices, up to 50 per page.

• LLM-ready Markdown of the page with include=markdown, or the raw JSON-LD blocks with include=jsonld.

• MCP-native with a free, keyless demo endpoint, so an agent can try it with one line of config.

Example (live, from the free demo):

{"name":"Men's Tree Runner","brand":"Allbirds","price":100,"currency":"USD","in_stock":true,"price_source":"json-ld"}

It doesn't get past anti-bot pages (Amazon, Best Buy, REI) and doesn't run JavaScript. For those, fetch the HTML yourself and pass it in.

Related MCP server: defuddle-mcp

Try it free

• No key, right now: point any MCP client at https://product-extractor.mohamed-bahaa53000.workers.dev/demo/mcp (10 tool calls per IP per day).

• Free RapidAPI plan: 50 requests/month on the Basic plan, no card needed. Pro is $9.99 for 5,000.

• Using n8n? Get competitor price monitoring installed in your n8n for $149, done for you.

This is a hosted (remote) MCP server using Streamable HTTP. There's nothing to install or run locally. This repository holds only the documentation and the MCP Registry manifest; the server code is not open source.

Official MCP Registry name: dev.workers.mohamed-bahaa53000.product-extractor/product-page-extractor

Tools

• extract_product: url or html, plus an optional include (product,meta,markdown,jsonld). Data is read from schema.org JSON-LD, microdata and OpenGraph.

• extract_batch: up to 10 URLs in one call, with ok/error for each page (paid endpoint only).

Use cases

• Price and stock checks for agents doing competitor monitoring.

• Catalog enrichment: GTIN/MPN/brand for product matching.

• Turning a product page into Markdown for an LLM to read.

Examples

Ask your agent: "What's the price and stock status of https://www.allbirds.com/products/mens-tree-runners?" It calls:

{"name":"extract_product","arguments":{"url":"https://www.allbirds.com/products/mens-tree-runners"}}

Already have the HTML (from your own crawler, or a site that blocks bots)? Pass it in and skip the fetch:

{"name":"extract_product","arguments":{"html":"<html>...</html>","include":"product,markdown"}}

Plain REST works too:

curl "https://product-extractor.mohamed-bahaa53000.workers.dev/demo/v1/extract?url=https://www.allbirds.com/products/mens-tree-runners"

Connect

{
  "mcpServers": {
    "product-extractor": {
      "url": "https://product-extractor.mohamed-bahaa53000.workers.dev/demo/mcp"
    }
  }
}

Paid / production (RapidAPI):

{
  "mcpServers": {
    "product-extractor": {
      "url": "https://product-page-extractor.p.rapidapi.com/mcp",
      "headers": {
        "X-RapidAPI-Key": "YOUR_RAPIDAPI_KEY",
        "X-RapidAPI-Host": "product-page-extractor.p.rapidapi.com"
      }
    }
  }
}

For clients that only support stdio: npx -y mcp-remote https://product-extractor.mohamed-bahaa53000.workers.dev/demo/mcp

Notes

• Stateless. Pages and HTML aren't stored beyond standard request logs.

• Only fetches public http/https pages.

License

MIT for this repository's docs and config examples (see LICENSE). The hosted service is proprietary.