MicrosoftMicrosoft Certified: Multi-Agent AI Solutions Expert2026-102.3 Build and integrate tool ecosystems
Exercise - Create an MCP server with Azure Functions
Completed by opening the official lab
Why it matters
Azure Functions turns existing serverless handlers into MCP tools without a separate server process, which makes it a common first hosting choice.
What to practise
Configure host.json with the server name, version and instructions under mcpToolTrigger, using the Preview extension bundle that the trigger requires. Then define summarize_text and classify_document with @app.generic_trigger(), toolName, description and toolProperties. Run the app locally and connect GitHub Copilot in agent mode to it through .vscode/mcp.json. Watch whether Copilot chooses the right tool from a natural-language prompt. Everything runs locally; nothing is deployed or secured.
How it fits
The trigger, tool properties and host.json settings are 2.3.5. The exercise connects through the separate SSE endpoint /runtime/webhooks/mcp/sse; 2.3.5 recommends the Streamable HTTP endpoint /runtime/webhooks/mcp unless a client requires SSE, and 2.3.3 explains the difference. Endpoint access controls, which this exercise skips, are also 2.3.5.