Copilot browser tools in VS Code reach general availability

GitHub made browser tools for Copilot in Visual Studio Code generally available on July 1. Agents can open and navigate pages, interact with controls, capture screenshots and console errors, and run scripted browser flows. According to the GitHub changelog, the integrated browser also exposes developer tools for a person to inspect the page directly.

Browser state has explicit boundaries

Tabs opened by an agent run in fresh sessions without the cookies or local storage from everyday browsing. A developer’s existing tab remains private until it is explicitly shared with the agent, and that access can be revoked. GitHub says sensitive browser permissions such as camera, microphone, location, notifications, and clipboard reads require user approval for a site.

Those controls reduce accidental access, but a shared authenticated page can still expose valuable data and actions. Test automation should use purpose-built accounts and environments rather than a developer’s production session whenever possible.

Make web tasks observable and disposable

Start with a narrow flow that has a clear assertion, such as loading a preview, submitting a test form, and checking one resulting state. Keep seed data deterministic and reset it after the run. Capture console errors and screenshots as debugging artifacts, while filtering credentials, personal data, and session tokens from logs.

Enterprise administrators can centrally switch browser tools off and apply allowed and denied network domains. Review wildcard rules and remember that denied domains take precedence, as GitHub documents. Workspace trust and approval prompts also remain relevant. For consequential operations, require the agent to stop before submission and let a reviewer inspect both the target page and proposed values. Browser automation is most useful when its side effects are bounded and its outcome can be checked independently.