Minimal MCP server / client that connects Azure OpenAI with a browser-automation toolset via Playwright.
https://github.com/kimtth/mcp-aoai-web-browsingBuilding AI agents that can actually interact with websites? You know the drill: days of wiring up Playwright, debugging protocol conversions, and writing custom bridges between your AI model and browser automation. This MCP server cuts through all that complexity.
Most developers end up building the same infrastructure repeatedly - connecting LLMs to browser automation, handling protocol translations, managing browser sessions, and debugging why their AI agent can't click that login button. You're not building your core application; you're rebuilding the same integration layer everyone else has already built.
Direct Azure OpenAI → Browser Connection: No custom API wrappers or protocol translations. Your Azure OpenAI deployment talks directly to Playwright through a clean MCP interface that handles all the messy conversion work.
Working Client/Server Implementation: Clone, configure your Azure credentials, and you have a functioning AI agent that can navigate websites, fill forms, and extract data. The example client shows real browser automation - not just theory.
FastMCP + MCP-LLM Bridge Integration: Built on proven components that handle the heavy lifting. The server passes objects directly to the bridge for stable connections, avoiding the typical MCP-to-OpenAI integration headaches.
QA Automation with AI Insight: Your AI agent navigates test environments, fills forms with realistic data, and provides intelligent analysis of what it encounters - not just pass/fail results.
Dynamic Data Extraction: Point your agent at competitor websites, internal dashboards, or partner portals. It adapts to layout changes and extracts the data you need without brittle selectors.
User Journey Testing: Describe complex user flows in natural language. Your agent executes multi-step scenarios while providing contextual feedback about UX issues it discovers.
The server architecture makes it straightforward to embed into existing applications:
# Your existing Azure OpenAI setup works unchanged
# The MCP server handles browser commands through function calls
# Example from the repo shows complete navigation flows
Configure your Azure OpenAI credentials once, and your AI models inherit full browser automation capabilities. The bridge component translates MCP tool responses into OpenAI function calling format automatically.
Development Setup: Uses uv for dependency management (the current MCP CLI standard), includes debugger configuration for VS Code, and provides realistic examples you can modify for your use cases.
Production Ready: Built with FastMCP for performance and includes proper error handling for browser automation edge cases.
You're not writing another Playwright wrapper - you're getting a complete MCP server implementation that other developers have tested and refined. The bridge handles protocol nuances you'd spend weeks debugging, and the example client demonstrates patterns you can apply immediately.
The focus stays on your application logic instead of infrastructure plumbing. Configure your Azure OpenAI deployment, describe what you want your agent to do, and watch it execute complex browser interactions with the full context of what it's seeing and doing.
This is the browser automation integration you'd build eventually - but it's ready now, tested by other developers, and designed specifically for the Azure OpenAI + MCP ecosystem.