Playwright Sample MCP Workspace

This workspace contains related projects for remote browser automation with Microsoft Playwright Workspaces.

Projects

Project	Purpose	Runtime
azure-playwright-service-mcp	MCP server that creates and ends remote Chromium browser sessions through Azure Playwright Service.	Node.js
browser-automation-agent	Long-running command-line agent that uses the MCP server plus Browser Use CLI to automate the remote browser turn by turn.	Python
browser-automation-agent-v2	Long-running command-line agent built with Microsoft Agent Framework, Agent Framework skills, the MCP server, and a generic shell tool for Browser Use CLI automation.	Python
browser-automation-agent-v2-foundry	Foundry-hosted container version of the v2 agent, exposing the Responses protocol through Microsoft Agent Framework hosting.	Python + Docker

azure-playwright-service-mcp

This project exposes a Model Context Protocol server with two tools:

• create_browser_session
• end_browser_session

Both tools accept a sessionId. The server converts the configured Playwright Service WebSocket endpoint into the HTTPS provisioning endpoint, passes runId=<sessionId>, and extracts the returned sessionUrl CDP WebSocket URL.

The MCP server is stateless: it uses the same sessionId/runId to create or resolve a browser session, then uses CDP Browser.close when ending a session.

See azure-playwright-service-mcp\README.md for configuration and MCP client usage.

browser-automation-agent

This project is a command-line browser automation agent built with bu-agent-sdk.

It:

• Starts the local azure-playwright-service-mcp server over stdio when it needs MCP tools.
• Creates a Playwright Service browser session through MCP.
• Attaches Browser Use CLI to the returned CDP URL once.
• Keeps the Browser Use session alive across turns.
• Closes Browser Use and then calls MCP end_browser_session when the user asks to end the browser session.

The agent supports Azure OpenAI through Entra auth with DefaultAzureCredential and the Azure AI Foundry/OpenAI v1 Responses API.

See browser-automation-agent\README.md for setup and run instructions.

browser-automation-agent-v2

This project is a command-line browser automation agent built with Microsoft Agent Framework.

It:

• Uses FoundryChatClient with DefaultAzureCredential for Azure AI Foundry model access.
• Registers the local azure-playwright-service-mcp server through Agent Framework MCPStdioTool.
• Provides a generic run_shell tool that the agent uses to run Browser Use CLI commands.
• Loads an Agent Framework skill adapted from the Browser Use CLI skill for Azure Playwright Service remote browser sessions.
• Logs skill/MCP session creation in blue and run_shell/session cleanup in yellow.

See browser-automation-agent-v2\README.md for setup and run instructions.

browser-automation-agent-v2-foundry

This project packages the v2 Microsoft Agent Framework agent as a Foundry hosted agent.

It:

• Runs ResponsesHostServer on port 8088 inside a container.
• Embeds the Azure Playwright Service MCP server so the hosted agent is self-contained.
• Uses the same Browser Use CLI skill and run_shell flow as browser-automation-agent-v2.
• Deploys with azd ai agent init and azd deploy.

See browser-automation-agent-v2-foundry\README.md for local hosting and Foundry deployment instructions.

High-level flow

User
  -> browser-automation-agent, browser-automation-agent-v2, or browser-automation-agent-v2-foundry
      -> azure-playwright-service-mcp
          -> Azure Playwright Service
      -> browser-use CLI
          -> remote browser CDP session

The MCP server handles browser session provisioning and cleanup. The agent handles the interactive automation loop and delegates concrete browser actions to Browser Use CLI.