AI Agents · TRACK 03/05
Model Context Protocol (MCP)
The USB-C of AI tools: using, building, and securing MCP servers.
// THE TRACK
01 · START HEREMCP (Model Context Protocol)Understand the open standard that lets any AI app plug into any tool or data source without custom integration code.BEGINNERBuilding MCP ServersWalk through every layer of an MCP server — primitives, SDK setup, tool and resource definitions, transport choices, and how to wire it into Claude.INTERMEDIATEConnecting Your First ServerGo from zero to a working MCP connection: install a server, point your AI app at it, and watch the model call its first MCP tool.BEGINNERBuild Your Own ServerWrite a small but real MCP server with the official SDKs, expose a tool, and connect it to a live AI client end to end.INTERMEDIATETools, Resources & PromptsLearn what each MCP primitive is for, who controls it (model, app, or user), and how to decide which one your feature should be.INTERMEDIATETransports: stdio vs HTTPUnderstand how MCP messages actually move — local subprocesses over stdio versus remote servers over Streamable HTTP — and when to pick each.INTERMEDIATEMCP vs Function CallingCut through the confusion: function calling is the model-side mechanism, MCP is the integration layer — learn when each is the right choice.INTERMEDIATEMCP Security RisksUnderstand the real attack surface MCP opens up — poisoned tools, injected instructions, and over-broad permissions — and how to vet a server before you trust it.INTERMEDIATEMCP Authorization & OAuthLearn how remote MCP servers authenticate users with OAuth, what scopes and tokens mean here, and why this is what makes hosted MCP usable safely.INTERMEDIATESampling, Roots & ElicitationGo past tools and resources to the MCP features people miss — sampling, roots, and elicitation — and learn what each one unlocks.INTERMEDIATE