mcp-aoai-web-browsing

This MCP server provides a bridge to Azure OpenAI, exposing a web-browsing-capable toolset via the MCP protocol. It uses Playwright under the hood to navigate web pages and interfaces with Azure OpenAI to produce function-call-like outputs compatible with OpenAI's function calling format. The server is implemented with FastMCP and integrates an MCP-LLM bridge so that MCP responses describing tools can be consumed by an OpenAI-compatible LLM. To use it, run the server via UV in Python, configure your Azure OpenAI credentials, and invoke the exposed MCP tools (for example, navigating to a URL through the provided Playwright-based tool). This setup enables secure, local browsing interactions driven by an LLM.

Prerequisites:

• Python installed on your system
• UV installed for Python dependency management (pip install uv)
• Azure OpenAI resources (endpoint, API key, deployment model) prior to running

Step-by-step:

1. Clone or download the MCP server repository.

2. Rename the example environment template if present and create a real .env file, or directly set the environment variables (see env placeholders in this guide): AZURE_OPEN_AI_ENDPOINT= AZURE_OPEN_AI_API_KEY= AZURE_OPEN_AI_DEPLOYMENT_MODEL= AZURE_OPEN_AI_API_VERSION=

3. Install UV and sync dependencies: pip install uv uv sync

4. Start the MCP server using UV (this runs the specified Python script): uv run chatgui.py

5. Ensure the server starts successfully and the client UI is accessible. If you see browser-related errors, you may need to install or configure Playwright dependencies as prompted by UV/Playwright integration.

Notes:

• The README describes running via uv run chatgui.py; adjust the script name if your entrypoint differs.
• Keep Azure credentials secure and do not commit them to source control.

Tips and common notes:

• Make sure to rename or create the .env file from the template and fill in Azure OpenAI details before starting the server.
• The MCP bridge converts MCP tool descriptions to OpenAI function calling formats, so ensure your LLM prompts expect this structure.
• If the Python process needs to be restarted, you can use the provided taskkill command for Python on Windows: taskkill /IM python.exe /F
• Playwright is used for browser automation; ensure your environment supports launching a browser (headless or headed) depending on your debugging needs.
• If you modify dependencies, run uv sync to install any new requirements.
• The server relies on the MCP protocol; verify your client configuration aligns with MCP tool outputs (names, descriptions, and input schemas).
• If you encounter environment issues, check that the Azure API version and deployment model match what your Azure OpenAI resource expects.