Skip to main content
Most AI web tools hand your AI a summary of a page. Something else read the page first and decided what mattered. If it skipped the part you needed, your AI never knows. Arora’s web reader brings back the actual page, the whole thing. Long pages come back in order, one section at a time, so nothing in the middle gets dropped.

​
Reads everything, without flooding your chat

Reading the whole page sounds expensive. It isn’t, because your AI doesn’t have to hold the whole page at once. Take the aroramcp.com homepage. The page is 75 KB. Arora’s first reply is under 3 KB: a short preview plus the page’s outline. That’s 96% less for your AI to take in up front, which saves tokens and time. And because it still knows every section that’s there, it reads the exact part your question needs, so answers come from the page itself, not a guess.
  • Big pages wait on your workspace. Your AI gets a short preview and the page’s outline, then reads only the sections your question needs.
  • Your AI sees the size first. Every result says roughly how long the page is before anything big gets loaded, so your AI can plan.
  • It takes just the slice you need. For API specs, your AI can pull one endpoint or one data type instead of the whole file.
  • It points to cleaner copies. When a site publishes a plain-text version of a page, Arora tells your AI so it can read that instead.
  • It skips empty pages. On JavaScript-built pages, your AI gets the product data hidden inside, not the empty page around it.
You get the full page when you need it and only the part you need the rest of the time. Your context window lasts longer, and long chats stay sharp.

​
Big pages

When a page is too big to fit in one reply, Arora puts the whole thing on your workspace and gives your AI an outline. Your AI then reads just the sections it needs. Nothing gets cut off.

​
API specs

For machine-readable specs, your AI can ask for just one slice, like a single endpoint or a single data type, along with every definition it depends on. Useful when you’re mapping fields between two systems.

​
What it can’t read

  • Pages behind a login. The reader doesn’t send your passwords or cookies. If you have to sign in to see it, Arora can’t read it this way. Use a connector for that app instead.
  • Non-HTTPS addresses. The link must start with https://.
  • Pages built entirely by JavaScript. Some sites send an empty shell and fill it in later in the browser. When that happens, Arora looks for the structured product and page data many sites include for search engines and hands your AI that instead. If there’s nothing to recover, Arora says so rather than pretending the page was empty.

​
Limits

Tool name: fetch_url. Read-only. Supports a select dot-path to return part of a JSON document, and expand_refs to resolve the $ref schemas that slice depends on (100 by default, up to 500). Long documents page with next_offset.
Last modified on October 13, 2026