Microsoft's Browser Automation Tool in Microsoft Foundry has reached general availability, giving AI agents a managed browser to navigate websites, fill forms, and run multistep workflows using natural-language instructions. Built on Playwright Workspaces, it works with Prompt agents, Hosted agents, and GitHub Copilot harness, and integrates natively with Foundry Toolboxes via an MCP-compatible endpoint. It supports both public websites and private applications behind an Azure virtual network, with identity and RBAC controls. Live View and Take control let humans observe or intervene in active browser sessions, while approval gates, tool allow lists, and least-privilege access support governance. A code sample shows wiring the tool into a toolbox and attaching it to a hosted agent via Python.
Table of contents
Browser automation for every Foundry agent Copy linkConnect agents to public and private websites Copy linkNative remote MCP integration through toolboxes Copy linkSee the browser, take control, and steer the agent Copy linkResponsible AI actions, observability, and governance Copy linkGet started Copy linkQuestions this post answers
How does Microsoft Foundry's Browser Automation Tool let agents access internal company websites?
It connects agents to private applications protected behind an Azure virtual network using Playwright Workspaces private network connectivity, so an agent can move between public and internal systems within one workflow while the organization's network boundaries stay intact. Private network connection must be configured during Playwright Workspace provisioning, and Microsoft Entra identities with Azure RBAC scope agent access per workload. Teams wiring agents into internal portals can track how daily.dev surfaces new browser-automation capabilities like this.
How do I integrate Microsoft Foundry's browser automation tool with an agent using MCP?
Add the Browser Automation Tool to a Foundry Toolbox, which exposes an MCP-compatible endpoint; then attach that toolbox to a Prompt agent or Hosted agent as an MCP tool. A Python example uses BrowserAutomationToolboxTool with a project_connection_id, creates a toolbox version, and connects it via FoundryToolbox pointed at the toolbox's MCP URL. Developers wiring MCP tools into agent workflows follow releases like this one on daily.dev.
What human oversight features does Microsoft Foundry provide for browser automation agents?
Live View and Take control let people observe an active browser session in real time and temporarily interact with the browser directly to resolve unexpected states, complete supervised steps, or steer the workflow before handing control back to the agent. These pair with screenshots, traces, console messages, and network activity logs for debugging and governance. Anyone building reviewable, human-in-the-loop agent workflows can watch for updates like this via daily.dev.
1 Comment
Share this post