MCP Servers Beyond 101: Good Practices, Design Choices and Their Consequences

The Model Context Protocol (MCP) has quickly become the connective tissue between Large Language Models and the tools, APIs, and data they rely on. But building a working MCP server is only the start. The real challenge comes in the design choices: generic vs domain-specific, open vs restricted, flexible vs safe.
In this talk from-the-trenches, we’ll go beyond the 101 level. Drawing from real-world projects, we’ll explore the patterns, pitfalls, and best practices that are emerging as MCP adoption accelerates. Through extended demos, architecture discussions, and practical examples, we’ll look at how different design decisions shape developer experience, performance, and security.
We’ll revisit lessons from the early days of REST APIs and see how they apply to MCP today—while recognizing that the stakes are higher now, since LLMs and agents are acting directly on our systems. By the end, you’ll have a clear mental framework to design smarter MCP servers, and a set of practical guidelines you can use immediately in your own projects.