MCP in Practice

How tools actually connect to models: where the seams are, how authority and scope are decided, and what it takes to wire it up without surprises.

For engineers connecting models to real tools and data — not demo-only wrappers.

9 parts. Start at Part 1.

  1. Part 1 — Why Connecting AI to Real Systems Is Still Hard

    Beginner

    Why connecting AI to real systems is still custom work: the N×M tax, frozen knowledge, and the missing standard.

  2. Part 2 — What MCP Is and How AI Agents Connect

    Beginner

    What MCP actually is, what it standardizes, and how it gives AI agents structured access to real systems.

  3. Part 3 — How MCP Works — The Complete Request Flow

    Intermediate

    The full MCP request flow — from client to server and back — with protocol-level detail.

  4. Part 4 — MCP vs Everything Else

    Intermediate

    MCP vs APIs, plugins, function calling, and agent tools — when to use each in real systems.

  5. Part 5 — Build Your First MCP Server (and Client)

    Intermediate

    A working MCP server and client from scratch, with the implementation decisions that matter.

  6. Part 6 — Your MCP Server Worked Locally. What Changes in Production?

    Intermediate

    One server, six stages: from local stdio prototype to deployed, authenticated infrastructure.

  7. Part 7 — MCP Transport and Auth in Practice

    Intermediate

    Two transports, three auth phases, and the deployment decisions for remote MCP servers.

  8. Part 8 — Your MCP Server Is Authenticated. It Is Not Safe Yet.

    Advanced

    Tool poisoning, rug pulls, and cross-server shadowing after transport and auth are in place.

  9. Part 9 — From Concepts to a Hands-On Example

    Intermediate

    The Part 5 order assistant moved from stdio to Streamable HTTP: same tools, different transport.