Model Context Protocol

The Model Context Protocol is an open standard for connecting AI applications to external capabilities. A server exposes tools, resources, and prompt templates; a client — an assistant application — discovers and invokes them. The point is decoupling: a capability implemented once becomes available to any compliant client, rather than being rebuilt per assistant.

Structurally it is Tool Use with the definitions supplied at runtime by a separate process rather than compiled into the host. That inversion is what makes it composable, and it introduces the ordinary problems of a plugin system: discovery, versioning, authentication, and the fact that every connected server's tool definitions consume Context Window.

Two practical themes recur when building on it. Authentication is the hard part — deciding how an end user's credential reaches a server, and what that credential is scoped to, is more design work than the tools themselves. And every connected server is trusted input: its tool descriptions are text the model reads, which makes a hostile or compromised server a Prompt Injection vector.

The same architecture appears wherever an application platform wants to expose itself to agents — a content system registering its capabilities as discoverable abilities is the same idea under a different name.