GitHub Copilot browser tools are GA in VS Code

GitHub Copilot browser tools in VS Code are now generally available. Confidence level: confirmed. Agents can open and navigate live web apps, click, type, read page content, capture screenshots, inspect console errors, and feed the results back into chat while VS Code keeps user-opened tabs private unless shared.

What changed

GitHub moved browser tools for Copilot in VS Code from preview to general availability on July 1, 2026. Browser tools are now on by default with GA. The agent can work with a real browser session, inspect the page state, capture browser evidence, and run scripted flows when repeated tool calls would be slower.

The change matters most for web developers who already ask agents to build or fix UI. The agent can now test the running page inside VS Code instead of stopping at code edits. GitHub says DevTools remain available in the browser toolbar, so developers can still inspect elements and console output themselves.

Key takeaways

• Browser tools for Copilot in VS Code are generally available, not just a preview flag.
• Agent-opened tabs run in isolated sessions without access to everyday browsing cookies or storage.
• User-opened tabs stay private until the user selects Share with Agent.
• Camera, microphone, location, notifications, and clipboard reads require explicit approval.
• Enterprise admins can disable browser tools or restrict reachable domains.

Availability and access

The feature is available in VS Code and the Agents window after updating VS Code. GitHub says browser tools are on by default with general availability. The VS Code 1.127 release notes also list browser tools for agents as generally available, and the VS Code guide documents the available browser tool actions.

Admins get policy controls rather than a forced open browser surface. GitHub lists a dedicated workbench.browser.enableChatTools switch, plus chat.agent.allowedNetworkDomains and chat.agent.deniedNetworkDomains when chat.agent.networkFilter is enabled. Workspace trust and approval prompts still apply.

Practical LinkLoot angle

This is useful when the agent needs proof, not just code. A practical loop is: ask Copilot to edit a UI, let it open the local app, watch it click through the flow, then review the screenshot, console output, and patch before accepting changes. That can catch broken buttons, routing errors, form validation issues, and obvious layout failures earlier.

For security-sensitive work, start with a narrow domain allowlist and keep logged-in production systems out of shared tabs. The agent can test local and staging pages without getting broad access to personal browser state. Teams building AI-assisted web workflows can pair this with LinkLoot's AI workflow automation guide.

What to verify before you act

• Confirm your VS Code build includes the 1.127 browser-tools GA changes.
• Check whether your organization manages workbench.browser.enableChatTools.
• Review allowed and denied network domains before letting agents browse.
• Test on local or staging apps before using the feature around production admin surfaces.
• Keep human review on screenshots, console findings, and generated fixes.

Source check

Confirmed by GitHub: browser tools for Copilot in VS Code are generally available; agents can navigate, click, type, inspect console errors, take screenshots, and run scripted flows; tabs and permissions are separated between agent-opened and user-shared sessions.

Confirmed by VS Code documentation: browser agent tools include page navigation, page reading, screenshots, user interaction, dialog handling, and custom browser automation. The VS Code 1.127 release notes also list browser tools for agents as generally available.

Independent context: NT Compatible's VS Code 1.127 coverage also identifies GA browser tools and per-site permission handling as headline changes. Use it as context, not as the primary source.