Let Claude browse the internet for you.

Open any page, search the web, read PDFs and pull data, from claude.ai on any device. Nothing to install.

Add to Claude How it works

Free for 300 page loads a month. Sign in with Google. Works in the Claude web, desktop and phone apps.

Summarise the MCP article

Open https://en.wikipedia.org/wiki/Model_Context_Protocol and give me the five key points

BrowsingBee · open_page en.wikipedia.org

Here are the five key points from the article:

Model Context Protocol (MCP) is an open standard Anthropic introduced on 25 November 2024.
It gives AI assistants one way to connect to outside tools and data instead of a custom integration for each.
It uses a client–server design: the assistant is the client, each tool provider runs a server.
Servers expose tools, resources and prompts; messages travel as JSON-RPC.
Other AI companies adopted it during 2025, and security researchers have flagged prompt-injection risks when servers return untrusted content.

Read 1 page · 12,000 characters · 4 s

Reply to Claude…

Claude Opus 5.5

Works on your phone

No laptop needed.

Connect BrowsingBee once on claude.ai. From then on the Claude app on your phone can open pages, search and read PDFs for you, because the browser runs on our servers, not on your device.

Ask for a page summary while you are out
Compare two product pages without opening either
Read a PDF link someone sent you and get the deadlines

Add to Claude

9:41●●●

Browser Rendering free limit

What is the free-plan limit for Cloudflare Browser Rendering? Find the official page.

BrowsingBee · search_web cloudflare browser rendering limits

BrowsingBee · open_page developers.cloudflare.com

On the Workers Free plan you get 10 minutes of browser time per day. Past that you get a 429 error until the next UTC day. The Workers Paid plan raises the limit.

developers.cloudflare.com/browser-run/limits

Searched the web, then read 1 page

Reply to Claude…

Seven tools, all read only

What Claude can ask BrowsingBee to do.

open_page

Opens a page in a real browser and returns the readable text as Markdown, with its links.

search_web

Searches the web and returns titles, addresses and snippets to open next.

screenshot_page

A PNG of the page, first screen or full page, desktop or phone width.

extract_data

The page content shaped for pulling out prices, specs, tables or contacts.

read_pdf

Text of a public PDF, page by page, up to 50 pages.

get_links

Every link on a page, optionally filtered, to find the right sub-page.

usage

Your plan, page loads used and when the count resets.

What it does

Opens public pages in a real Chrome, so JavaScript-rendered sites work
Returns clean text, not raw HTML, so answers stay cheap and fast
Follows links, searches, screenshots and reads PDFs in one conversation
Respects robots.txt and identifies itself as BrowsingBee

What it doesn't do

Log in to sites, fill forms or buy anything
Open pages behind a login, a paywall or a CAPTCHA
Reach private networks or local addresses
Keep your page content: only a host name and a count are logged, screenshots expire in 24 hours

Heavy pages take up to 10 seconds. Slow is fine. Wrong is not.

Pricing