The Model Context Protocol (MCP) is emerging as the de-facto standard for connecting AI agents to tools, APIs, and enterprise systems — a “USB-C for AI” that promises to kill the M×N integration problem.

But production MCP isn’t about demos. It’s about authentication, client identity, and trust boundaries.

In this session, we do a code-first MCP deep dive focused on what teams actually struggle with: secure auth flows, client registration without unsafe dynamic onboarding, token scopes, and clear responsibility between MCP client, host, and server.

Using the Microsoft Agent Framework as an MCP client, we’ll show real patterns that work today — and the pitfalls you need to avoid before MCP hits production.

MCP is easy to demo — and hard to secure. This session is a code-first deep dive into MCP authentication and client design, using the Microsoft Agent Framework as a real MCP client.