Microsoft Foundry's browser automation tool is now generally available

The tool gives agents a managed browser to navigate websites, inspect content, complete forms and carry out multistep workflows. Agents work through an observe-and-act loop, checking the current page, choosing an operation, acting, and verifying the result. There is no need to install browser binaries or run browser infrastructure in the agent environment, since Playwright Workspaces provisions isolated cloud sessions on demand. The Playwright Workspace must be provisioned separately as an Azure resource.

Private network connectivity is configured when the Playwright Workspace is provisioned, letting an agent move between public and internal systems in one workflow. Access can be controlled with Microsoft Entra identities and Azure role-based access control. The tool is also integrated with Toolboxes in Foundry, so it can be added to a toolbox and connected to Prompt or Hosted agents through an MCP-compatible endpoint.

Live View lets operators watch an active session in real time, and Take control lets a person interact with the browser directly before handing the workflow back to the agent. Organizations can require approval for state-changing operations and use isolated sessions per workflow. Foundry traces may contain sensitive content, and the blog points to guidance on restricting access to it.

Agents can reach internal portals, legacy systems and vendor sites without custom APIs, using managed browsers instead of infrastructure they run themselves.

Developers can start with navigation and page-inspection tools and enable clicking, typing or file upload only when a scenario needs them. Approval controls and allow lists help limit what an agent can change.

Web content should be treated as untrusted input, so system policies should stay separate from page content and consequential actions should require human approval.