Real Intelligence: Developing effective AI services for humans with MCP.

“MCP” classes often use messy network interfaces in Python; they describe py* classes not MCP. This workshop will start with MCP structures that define a service, with alternatives, develop a local Claude extension, and return effective errors that correct the AI to just Make it Work (tm).

The MCP interface standard is not about Python or Javascript. Unfortunately, all the classes I’ve seen so far use examples of overly-complicated Python that involve network interfaces and don’t really describe the underlying MCP structure. This hands-on workshop will start with the MCP structures that describe an interface, look at alternatives for describing the service to the AI, add the extension to Claude, and incrementally extend the code to return effective errors that correct the AI’s mistakes, and build a simple, robust handler for natural language integration. We’ll also look at using Claude’s ‘haiku’ to get suggestions for extending the structures to develop a richer service. The underlying service can easily be extended to a network, but the point here is architecting the MCP structure, not the

This could go anywhere from a 1-hour lecture to a half-day sidebar, to a full class on Monday; I could easily give a one-hour session in the lectures and have the workshop before or after the main sessions. I’m really, really sick of seeing all the “AI” coding be Python as if that is the only way to do it when the MCP standard has nothing to do with Python, or coding.