GitHub Copilot browser tools go GA
Copilot agents in VS Code now navigate live web apps, click, type, take screenshots, and capture console errors, with Enterprise domain allowlist controls.
GitHub Copilot’s browser tools reached general availability on July 1 in VS Code. Copilot agents can now navigate live web applications, click elements, type text, hover, drag, handle dialogs, read page content, capture console errors, and take screenshots. Developer tabs stay private; agent activity runs in isolated agent tabs.
Enterprise controls ship with the GA release. A workbench.browser.enableChatTools workspace setting governs the feature globally. Admins can specify chat.agent.allowedNetworkDomains and chat.agent.deniedNetworkDomains to restrict which addresses the agent can reach. The feature is enabled by default for Pro, Pro+, and Max accounts.
The practical use case is tightening the UI feedback loop. A coding agent can open a pull request preview, interact with a form, and surface a console error without a separate Playwright or Selenium harness. For that loop to work reliably, the agent needs the same access the developer has, which means authenticated sessions can be in scope. The coding agent security guide covers what an agent with browser access can reach and what to gate behind a human approval step.
The domain allowlist controls are worth configuring for any team using this in a shared or regulated environment. The feature propagates through account tiers over the coming weeks, so the window to set policy before deployment is now. This is the same access-scoping discipline as giving your AI agents their own keys: define the boundary before handing over credentials.