As AI agents become a first-class part of software systems, learning to build MCP servers is becoming as fundamental as learning to build RESTful APIs. Many engineers already interact with MCP through IDEs and agent tooling, and building these servers introduces a new set of design considerations beyond traditional request/response APIs. While MCP servers share familiar API concepts, the move to agent-driven, probabilistic clients changes how engineers think about contracts, tool design, output shaping, state management, error handling, and spec adoption. Building MCP servers is emerging as an important capability for all developers and teams.

In this session, we’ll start with what MCP is and how the protocol defines resources, tools, prompts, and elicitations. We’ll walk through the practical decisions involved in building your first MCP server, including transport choices, authentication, tool and resource granularity, server instructions, and context engineering. You’ll see how MCP aligns with and diverges from traditional API design, explore framework options, preview generated MCP servers, and learn how to design for scaling, versioning, gateways, and deterministic error handling in real-world systems.