Native Anthropic CUA Support in Anchor Browser

Anthropic’s Computer Use API (CUA) represents a major shift in how AI agents interact with the web. Instead of relying on brittle DOM automation or site specific APIs, models can now see, reason about, and operate real browser environments like a human user would.

Anchor Browser now provides native support for Anthropic CUA, making it practical to run CUA agents in production environments with high reliability, speed, and scale.

Why Computer Use Needs Specialized Browser Infrastructure

CUA models are powerful, but browser automation remains unforgiving. Most real world failures are not caused by reasoning errors, but by infrastructure gaps:

Incomplete browser control
Bot detection and IP reputation issues
Slow or unstable runtimes
Poor recovery from edge cases

Anchor was built specifically to solve these problems for Agentic browser workloads.

What Makes Anchor the Best Runtime for Anthropic CUA

OS Level Browser Controls

Anchor operates at the operating system level rather than just the DOM. This gives CUA agents full control over real Chromium instances, including:

Native mouse and keyboard interactions
Window focus and scrolling behavior
File dialogs and downloads
Accurate rendering and viewport behavior

This matches the mental model CUA uses when reasoning visually about the screen.

Custom Tools for Direct Browser Access

While CUA is fundamentally visual, there are moments where structured access is needed. Anchor exposes custom tools that allow agents to:

Read browser state deterministically
Execute precise actions when visual interaction is inefficient
Recover cleanly from edge cases

This hybrid approach combines the flexibility of visual reasoning with the reliability of direct browser primitives.

Humanized Browser Behavior and Reliable IPs

Anchor sessions are designed to behave like real users:

Human like interaction timing and input patterns
Stable, high reputation IP addresses
Stealth hardened Chromium configurations via Anchor Chromium

This significantly reduces blocks, CAPTCHAs, and flaky behavior, especially for long running CUA workflows on production websites.

Fast Runtime

Anchor is optimized for low latency execution:

Fast browser startup
Minimal overhead per action

CUA agents complete workflows faster, with fewer timeouts and higher overall success rates.

Using Anthropic CUA with Anchor

Below is a minimal example showing how to run an Anthropic CUA agent inside an Anchor Browser session using Playwright over CDP.

‍

import { chromium } from 'playwright';

const browser = await chromium.connectOverCDP(connectionString);
const context = browser.contexts()[0];
const ai = context.serviceWorkers()[0];

const result = await ai.evaluate(JSON.stringify({
  prompt: "Go to el al website and find flights to new york from israel tomorrow.",
  agent: "anthropic-cua",
  model: "claude-opus-4-5-20251101"
}));

console.info(result);

‍

Anchor manages the browser lifecycle, OS level control, IP routing, and performance characteristics. Anthropic CUA focuses purely on perception, reasoning, and decision making.

Production Ready Computer Use

Anthropic CUA unlocks a new class of automation, but only when paired with infrastructure designed for real world web execution.

Anchor gives Anthropic CUA agents a browser environment that behaves like a real human, runs fast, and holds up under production workloads.

If you are building serious browser agents with Anthropic CUA, this is the runtime they were designed for.

‍