Browser Use Brings 84.8% Stealth Web Browsing to Anthropic's Claude Agents

Browser Use's stealth cloud browser is now a one-package install inside Anthropic's Claude Managed Agents, giving hosted agents live web access with CAPTCHA solving and residential proxies at $0.02/hr.

·
·
  • Browser Use is now available inside Claude Managed Agents, giving hosted Claude agents live web access via a stealth cloud browser.
  • Top stealth scores: Browser Use Cloud leads all providers on BrowserBench at 84.8% and on its own benchmark at 81%, nearly double Browserbase's 42%.
  • Five-step YAML setup installs the browser-use CLI in the agent environment; API key is stored in a vault and never exposed to the model.
  • Pricing starts at $0.02-$0.06/hr for cloud browser sessions with a proportional refund for unused time; $10 in free credits on signup.
  • Features include CAPTCHA solving, residential proxies, sub-second cold starts, live view, and mp4 session recording.
  • Claude Managed Agents is in public beta and now supports scheduled deployments, letting agents run recurring tasks like nightly syncs or weekly scans automatically.

The hardest part of building a browser agent has never been the AI. It's been the web itself: bot detection, CAPTCHAs, anti-scraping shields, and the sheer pain of provisioning a headless browser in a sandboxed cloud environment that has no display. Browser Use just solved that last piece for Claude users. The team announced that their stealth cloud browser is now available as a first-class integration inside Anthropic's Claude Managed Agents platform.

The problem this actually solves

Claude Managed Agents provides the harness and infrastructure for running Claude as an autonomous agent. Instead of building your own agent loop, tool execution, and runtime, you get a fully managed environment where Claude can read files, run commands, browse the web, and execute code securely. The catch? The sandbox can't run a local browser, so the agent has to connect to Browser Use Cloud with browser-use cloud connect. There's no display, no GUI, no Chromium to spin up locally. Until now, that meant web tasks were either off-limits or required awkward workarounds.

The bottleneck for agents is not their intelligence. On most websites users care about, agents get blocked by a CAPTCHA or an antibot challenge. Browser Use collected 300,000 security check events from production traffic, and the data made it clear: agent intelligence and browser stealth are orthogonal. You need to measure them separately.

What Browser Use brings to the table

Browser Use Cloud is not just a remote Chromium instance. It combines a purpose-built agent with their own browser infrastructure: stealth proxies, CAPTCHA solving, persistent filesystem, and optimized tool orchestration. The stealth numbers back this up. On the independent BrowserBench evaluation by Halluminate, which runs 296 tasks across antibot-protected sites, Browser Use Cloud achieved the highest success rate of all providers tested at 84.8%.

On Browser Use's own stealth benchmark, which is built from real production data across 71 sites protected by Cloudflare, Akamai, PerimeterX, and Datadome:

  • Browser Use Cloud: 81% (nearly double Browserbase's 42%)
  • Anchor: 77%
  • Onkernel: 67%
  • Steel: 47%
  • Hyperbrowser: 40%

Browser Use Cloud's highest bypass rates are against Cloudflare (93%), Akamai (85%), and PerimeterX (81%). These aren't soft targets.

Getting set up in five steps

The integration is a YAML-first setup that takes minutes. Here's the full flow:

  1. Create an environment that pre-installs the browser-use CLI and allows outbound connections to *.browser-use.com
  2. Create a credential vault storing your BROWSER_USE_API_KEY as an environment variable so the model never sees it directly
  3. Create the agent with a system prompt instructing it to always run browser-use cloud connect first before any navigation
  4. Start a session by sending a user.message event to the Managed Agents API with the managed-agents-2026-04-01 beta header
  5. Watch it run via live view and mp4 recording in the Browser Use Cloud dashboard

The agent system prompt is the key piece. It needs to know the exact CLI command sequence:

browser-use cloud connect   # provision the remote browser
browser-use open <url>      # navigate
browser-use state            # inspect clickable elements
browser-use click <i>       # interact
browser-use type "<text>"   # fill forms
browser-use eval "<js>"     # run arbitrary JavaScript

The agent config itself looks like this:

name: browser agent
model:
  id: claude-opus-4-8
system: |
  You are a browser agent. Use the browser-use CLI to complete web tasks.
  Always run browser-use cloud connect first to provision a remote browser.
tools:
  - type: agent_toolset_20260401
    default_config:
      enabled: true
      permission_policy:
        type: always_allow

What it costs

The Browser Use Cloud browser sessions are billed separately from your Claude API tokens. Browser sessions cost $0.06 per hour on the pay-as-you-go plan, with proxy data at $10/GB. The tweet quotes $0.02/hr, which likely reflects a discounted rate for the Managed Agents integration specifically. Browser Use starts with $10 in free credits when you sign up. Billing is upfront based on timeout duration, with a proportional refund for unused time when you stop the session.

For heavier workloads, the Starter plan at $100/month includes $100 in credits with discounted proxy rates ($7.50/GB) and up to 50 concurrent sessions. The Business plan at $500/month drops session costs to $0.03/hour and proxy data to $5/GB, with up to 250 concurrent sessions.

Where this fits in the broader picture

Claude Managed Agents is currently in beta. Sessions are long-running, resume cleanly after pauses, and store conversation history, sandbox state, and outputs server-side. Browser Use is not the only player adding browser capabilities to this platform: Browserbase and KERNEL also give Managed Agents browser capabilities, so agents can navigate and interact with the web alongside their other tools. And Cloudflare's integration with Claude Managed Agents lets developers run agent logic on the Claude platform while using Cloudflare for code execution, secure connectivity, and custom tools.

What makes the Browser Use angle distinct is the stealth-first architecture. Remote, CDP-compatible stealth browsers for AI agents with residential proxies, CAPTCHA solving, and benchmark-leading anti-detection, with sub-second cold starts. For tasks that involve real production websites, that difference between 84.8% and 42% task completion is the difference between a working product and a broken one.

The practical use cases are broad: competitive intelligence scraping, automated form submissions, login-gated data extraction, monitoring dashboards that require authentication, and any long-running async workflow where you need a Claude agent to interact with the live web without you babysitting it. Agents can now also run on a schedule, completing routine work automatically. A scheduled deployment gives an agent a cron schedule, and each time the schedule fires, the agent starts a new session and completes its task, with no scheduler for you to build or host. Use it for recurring work like a nightly data sync, a weekly compliance scan, or a daily digest.

Comments

avatar