MCP primitives: the mental model behind the protocol

MCP (Model Context Protocol) is built around five core primitives that define how agents interact with data and actions. Resources expose read-only context, tools provide executable actions with side effects, prompts offer reusable instructions, sampling controls model generation behavior, and roots define access boundaries. This separation keeps context, instruction, behavior, and execution distinct, making MCP systems more predictable and debuggable as they scale.