How to use agent.as_mcp_server() in Microsoft Agent Framework to expose a ChatAgent over the Model Context Protocol using stdio transport.

A deep dive into how agent.as_mcp_server() differs architecturally from a raw MCP SDK implementation — and why that difference matters.

How to run an MAF ChatAgent as a standalone HTTP/SSE MCP server and connect to it from a fully independent client process.

How to externalise agent configuration into YAML files with Pydantic validation, and load multiple agents at startup without touching Python code.

How to give a single MAF agent three simultaneous interfaces — MCP for AI clients, REST for traditional services, and a registry for service discovery — all derived from one tools list with zero duplication.