What Is Model Context Protocol (MCP) and Why Should You Care? A Beginner's Guide

In this recorded Live! 360 session, Fabian Williams introduces the Model Context Protocol (MCP) and shows how developers can use it to build agentic applications that work reliably across tools, clients, and models. You’ll learn why MCP is quickly becoming the “USB for AI,” how it standardizes tool-calling, and why separating tools, resources, and prompts gives you more control over security, privacy, and scale.

Fabian walks through real examples using Claude Desktop, ChatGPT, Copilot Studio, LM Studio, and VS Code—demonstrating how MCP servers operate locally or in the cloud, how grounding differs from model training, and how observability and guardrails fit into a production-ready agentic architecture. You’ll also see how MCP supports multi-model development and how organizations can manage agent behavior, governance, and telemetry at scale.

🔑 What You’ll Learn
• What MCP is and why it’s becoming the standard for agentic development
• How tools, resources, and prompts work together in an MCP server
• The difference between grounding vs. training an LLM
• When to run MCP locally (privacy, cost, air-gap) vs. hosted in the cloud
• How AI clients (Copilot, Claude, ChatGPT, LM Studio) discover and use MCP tools
• Key security considerations: prompt injection, tool exposure, permissions
• How observability, telemetry, and evals help monitor and govern agents
• Why small language models (SLMs) are critical for private and offline agentic workloads

⏱️ Chapters
00:00 Welcome, speaker introductions & session setup
02:42 How the demos work (local vs. cloud MCP, models, caveats)
06:59 Why MCP exists & the problem it solves
09:00 LLM limits, context windows & grounding
11:50 What MCP is: tools, resources & prompt flow
17:03 MCP across clients: Claude, ChatGPT & VS Code
21:45 Copilot Studio + MCP: connectors, tools & enterprise workflow
23:14 Running MCP locally vs. hosted
31:50 Demo: querying conference sessions with MCP
37:00 Guardrails, system prompts & preventing misuse
41:00 MCP vs. A2A (Agent to Agent), ACP (Agent Communication Protocol) & Claude Skills
43:57 Agent 365: Observability, OpenTelemetry & monitoring agents
48:00 Metrics, usage insights & multi-agent patterns
55:23 Demo: adding sessions to calendar using Microsoft Graph

👤 Speaker: Fabian Williams

🔗 Links
• Download Visual Studio 2026: http://visualstudio.com/download
• Visual Studio Live! events: https://aka.ms/VSLiveEvents

#ai #copilot