networks//network protocols//MCP

MCP (the Model Context Protocol) is a protocol that gives an AI application a shared interface for discovering and using capabilities exposed by another process. A client and server exchange structured messages, so each integration does not need a wholly separate tool-discovery convention. The interface describes how to request an operation; it does not make that operation trustworthy.

MCP (the Model Context Protocol) is a protocol that gives an AI application a shared interface for discovering and using capabilities exposed by another process. A client and server exchange structured messages, so each integration does not need a wholly separate tool-discovery convention. The interface describes how to request an operation; it does not make that operation trustworthy.

It came at the end of a short sequence. Language models first answered from their weights alone; then applications wired tools to them one integration at a time (tool use); Anthropic published MCP in November 2024 so that a tool written once as an MCP server works in any host that speaks the protocol (an assistant, an editor, an agent). The roles are host (the application running the model), client (one connection inside the host) and server (the program exposing tools, resources and prompts). What it is matters as much as what it is not: it is neither the model nor a way of training one, and it sits in front of the services' own APIs, which still need their credentials and permissions.

MCP before and after develops the integration pattern. Minimum privilege remains relevant because a standard interface can expose powerful actions just as easily as harmless ones.