Breaking

Anthropic Adds Model-Controlled In-App Browser to Claude Code

July 10, 2026 at 16:45 EDT

AI Agents
Software Dev & Coding

Anthropic has added a sandboxed in-app browser that Claude itself can control to the Claude Code desktop app. Claude can open documents, designs and any website, read them, and click through and interact with them the same way it handles local dev servers.

Anthropic · Claude Code

A Browser Claude Can Drive Itself

The Claude Code desktop app now ships with a sandboxed in-app browser the agent controls on its own — opening pages, reading documents, clicking through and filling out forms, no copy-and-paste required.

100%

Agent-driven browsing — Claude loads, clicks and fills forms autonomously

Manual copy-and-paste needed to review external site content

MCP

External tools and live data connect via Model Context Protocol

From manual review to a self-running loop

How the agent now handles a web task end to end

Open page / doc

→

Read & interact

→

Click / fill forms

→

Self-verify result

Everything runs in a sandbox

Native sandbox integration — not a browser plugin — is the differentiator.

Sandboxed actions

OS-level limits like Apple Events blocking, adjustable in settings

Session control

Users choose whether to persist state such as login

Combined isolation

Works with existing filesystem / network isolation and screen control

Developers praise

No more copy-and-paste to review sites
Claude opens docs and running apps to verify itself
Sandbox + session controls feel reassuring
"Getting closer to a real operator"

Still rough edges

resize_window can duplicate the viewport (fix: reload or resize the real window)
Feature gaps between desktop and terminal versions

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.