Daily AI digest

20 October 2026
Cloud computing and AI

Microsoft Foundry’s browser automation tool reaches general availability

Reported by Microsoft Foundry Blog◆Filed 20 October 2026
Editorial cartoon accompanying “Microsoft Foundry’s browser automation tool reaches general availability”.
Editorial illustration

The brief

Microsoft has made its Browser Automation Tool for Microsoft Foundry generally available, allowing agents to navigate websites, inspect pages, complete forms, and perform multistep workflows through managed cloud browser sessions. The tool supports public and private applications, remote MCP integration, observability, and human intervention, but requires a separately provisioned Playwright Workspace and safeguards for sensitive or state-changing actions.

Key points

  • The tool uses Playwright Workspaces to provision isolated browser sessions without requiring agents to install or manage browser infrastructure.
  • It supports public websites and private applications connected through Azure virtual networks, with Microsoft Entra identities and Azure RBAC for access control.
  • Developers can expose the tool through Foundry Toolboxes and connect it to Prompt agents, Hosted agents, or GitHub Copilot harness workflows using MCP-compatible integration.
  • Live View and Take control allow operators to observe active sessions and temporarily intervene when workflows encounter unexpected states or require human judgment.
  • Foundry traces may contain sensitive data; Microsoft’s documented protection workflow for routing such content to a restricted AppGenAIContent table is currently a preview capability.

Read the source report

Microsoft Foundry Blog ↗

Full summary

Microsoft has announced the general availability of Browser Automation Tool in Microsoft Foundry. The tool gives AI agents a managed browser for navigating websites, inspecting content, completing forms, and carrying out multistep workflows, including processes that rely on internal portals, vendor systems, dashboards, or applications without complete API coverage.

The tool is powered by Playwright Workspaces, which provisions isolated cloud browser sessions on demand and manages the browser infrastructure separately from the agent environment. It supports Prompt agents, Hosted agents built with Microsoft Agent Framework or other frameworks, and GitHub Copilot harness workflows. Agents use an observe-and-act process to inspect pages, perform browser operations, and verify resulting states as sites change or display dialogs and dynamic content.

Browser Automation Tool can access public websites and private applications protected behind an Azure virtual network. This enables workflows spanning internal and external systems, such as employee onboarding, vendor-data transfer, customer support investigations, and legacy applications without modern APIs. Private network connectivity must be configured when provisioning the Playwright Workspace, while Microsoft Entra identities and Azure role-based access control can limit agent permissions.

The tool is integrated with Microsoft Foundry Toolboxes, allowing developers to curate, govern, publish, and reuse browser automation configurations across agents through an MCP-compatible endpoint. Developers can begin with navigation and page-inspection capabilities and selectively enable actions such as clicking, typing, form filling, and file uploads. Live View provides real-time observation, while Take control lets a person intervene in an active session and then return control to the agent.

Microsoft highlights approval controls, tool allow lists, isolated sessions, least-privilege access, and tracing for governing browser actions that can affect real systems. It advises treating web content as untrusted input, separating system policies from page content, validating sensitive destinations, and requiring approval for consequential actions. Foundry traces can contain prompts, responses, instructions, and tool data; Microsoft’s guidance for restricting this information through the AppGenAIContent table and Azure RBAC is currently documented as a preview capability.