Extract structured data from any website with an API

The MCP server gives agents the same advice: look things up and read pages with web_search and web_fetch, and keep the browser agent for pages it has to interact with and sites no tool covers.

1. Check the catalog first

   Describe the job to POST /v1/tools/suggest or search with POST /v1/tools/discover. Both are free and return the tool, an example input and the cost. See Find the right tool.

2. Read or crawl public pages

   Run web-fetch on one URL or website-crawl on a section, and turn the Markdown into JSON with your parser or a model on the same key.

3. Extract with a schema when you must interact

   Start a browser agent run with a goal, a start URL and an output_schema. The agent clicks through and returns JSON that matches your schema in data.

This run opens a book store's Travel category and returns every book on the page as JSON:

curl https://lovelyrouter.com/v1/browser/runs \
  -H "Authorization: Bearer $LOVELY_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "goal": "Open the Travel category and list every book on the first page with its title, price, star rating, stock status and link.",
    "start_url": "https://books.toscrape.com/",
    "output_schema": {
      "books": [{"title": "string", "price": "price as shown, with currency", "rating": "star rating as shown", "in_stock": "boolean", "url": "string"}]
    }
  }'

The response is 202 with the run's id. Fetch GET /v1/browser/runs/{id}?wait=30 until the run finishes, then read data. The shorthand maps each field to a type; a description like "price as shown, with currency" makes the field a string the agent fills as you describe. For enums, formats and required fields, send a full JSON Schema (draft 2020-12) instead, up to 20 KB.

What makes the output safe to build on:

For public pages, a browser is more than you need. This crawl reads a careers section and stops at 50 pages:

curl https://lovelyrouter.com/v1/tools/website-crawl/run \
  -H "Authorization: Bearer $LOVELY_API_KEY" \
  -H "content-type: application/json" \
  -d '{
    "input": {
      "startUrls": [{"url": "https://www.example.com/careers/"}],
      "maxCrawlDepth": 2,
      "excludeUrlGlobs": ["https://www.example.com/careers/archive/**"]
    },
    "max_results": 50
  }'

Each item is one page as Markdown. Pass the pages to a model with the fields you want and you have the open roles as JSON. includeUrlGlobs widens the crawl beyond the start URL's sub-pages, and useSitemaps adds the URLs from the site's sitemap.

Reading a page costs $0.0009, and a page that can't be read costs nothing. The crawl, screenshot and contacts tools bill per page, screenshot or website, and the crawl adds a per-call fee; the cards above show today's prices. The browser agent bills per step, $0.006 in fast mode and $0.012 in accurate mode, with models and structured output included and no run fee. Every run holds its maximum cost first and settles to what it used. No subscription. See pricing.