Blog Post

Microsoft Foundry Blog
3 MIN READ

Introducing the New Browser Automation Tool with Toolboxes in Foundry

NandiniMuralidharan's avatar
Jun 03, 2026

Your AI agent just analyzed 500 invoices—but it can't click the 'Submit' button to process them. AI agents can reason through complex problems, but they hit a wall when workflows require interacting with web interfaces The Browser Automation Tool bridges that gap, giving your agents the ability to interact with any website the same way a human would. Here's what's new in our latest release with Toolboxes support in hosted agents.

Browser Automation Tool is available as an MCP (Model Context Protocol) tool powered by Playwright workspaces as its headless browser infrastructure layer. This gives teams a faster path from idea to automation, with better visibility into what is happening in real time and more control when workflows hit complex edge cases. If you’re working on web automation, this release is for you—it lets you use any CDP (Chrome Dev Tools Protocol) supported browser framework or reasoning SDK to automate your workflows with ease.

Why this release matters

Today’s AI agents are powerful at reasoning but fall short when workflows move beyond APIs.

Enterprise reality is:

  • Critical systems live behind web interfaces, not APIs
  • Workflows require navigation, decision-making, and multi-step interaction
  • Automation breaks at the last mile of execution
  • Private and internal websites are difficult to validate safely at scale.

The revised Browser Automation Tool with Toolboxes in Foundry is designed to close those gaps with a practical, operator-friendly workflow.

What is new

  1. MCP-native Browser Automation Tool in Toolboxes: Browser Automation Tool (Public Preview) is now available as an MCP tool in Toolboxes, working seamlessly with hosted agents in Foundry Agent Service. This makes it easier to adopt browser automation in your existing agentic workflows. Teams get standardized browser tasks with built-in Microsoft Entra ID authentication (formerly Azure AD).
  2. Playwright workspaces as the infrastructure layer: The underlying execution model is powered by Playwright workspaces (generally available) which brings robust browser automation foundations and reliable runtime behavior for modern web applications.
  3. Live View for issue detection: Live View (Public preview) helps customers detect issues as they happen. You can observe browser actions and quickly identify - selector drift (when UI elements change and automation can no longer find them), timing issues, navigation failures, unexpected UI states. This improves debugging velocity.
  4. Take Control for edge cases: When automation reaches a non-deterministic path (unpredictable user flows like CAPTCHAs or dynamic content), customers can use Take Control (Public preview) to intervene and steer execution through edge-case conditions. This blends automation scale with human judgment when it is needed most.
  5. Support for private website browsing [Private Preview]: Customers can browse and automate private websites in hosted environments, enabling more realistic enterprise workflows such as internal portals, authenticated research paths, and secured form flows. Playwright Workspaces Private Website Feature - Private Preview Enrollment
  6. Built-in Observability & Optimizations: Logs and evaluations are integrated within observability in Foundry Control Plane and full traceability is available for debugging and reliability
  7. Choose Orchestration layer: Customers can choose from wide variety of open-source reasoning layer for their use case while writing the hosted agents.


Get Started Here are the samples: Sample 1 (BYO) and Sample 2 (Microsoft Agent Framework)

Sample Scenarios could be Form filling (automate repetitive, high-volume form interactions while maintaining traceability and visibility during execution) or Web research (Scale browser-based research flows across multiple sources, while retaining the option to step in manually when pages or data structures vary)

How teams can use Browser Automation Tool in practice :

  • Define the desired browser workflow in your Hosted Agent setup.
  • Run Browser Automation Tool execution with Live View enabled.
  • Detect and resolve issues quickly as they appear.
  • Use Take Control on complex or unexpected branches.
  • Continue automation and capture outcomes for iteration.

What this unlocks

With hosted agents with Browser Automation Tool  , you can now build:

  • ✅ End-to-end agents that complete real workflows with reliable automation
  • ✅ Enterprise-grade systems with identity, compliance, and observability
  • ✅ Human-in-the-loop experiences when precision matters

This unlocks a new class of applications across sectors (finance & operations, back office automation, testing, web-based data workflows) where teams want to automate with confidence and not just speed.

Learn more

This update is part of a broader investment in practical, agent-driven browser automation. We are focused on making automation more adaptable to real-world website behavior. If your workflows depend on forms, web navigation, and research at scale, now is a great time to evaluate the revised Browser Automation Tool experience in hosted agents in Foundry Agent Service.

Updated Jun 07, 2026
Version 5.0
No CommentsBe the first to comment