The Model Context Protocol is an open standard for connecting AI applications to external tools and data sources. It defines one way for an assistant to discover what a system can do and to call it, so an integration written once works with any client that speaks the protocol.
Also called: MCP · Reviewed
Before a shared protocol, every AI client needed a bespoke integration with every tool: a dozen assistants and a dozen systems meant a hundred and forty-four separate pieces of glue, each with its own auth, its own schema and its own maintainer. MCP turns that into one server per system and one client implementation per assistant.
An MCP server sits in front of a system — a database, a file store, a ticketing tool, an internal API — and advertises what it offers. Tools are actions the model can invoke; resources are data it can read; prompts are reusable templates the server suggests. The client passes those descriptions to the model, and when the model asks to call one, the client routes the request to the server and returns the result.
Because the descriptions travel with the server, adding a capability does not require changing the assistant. The model discovers it.
Agents are only as useful as the things they can reach. A protocol that makes reaching a new system a matter of pointing at a server, rather than writing an adapter, is what lets an assistant be useful against your actual stack rather than a demo one.
It also draws a clean security boundary. The server decides what is exposed and with what permissions, which is a far better place to enforce least privilege than inside a prompt. Treat anything an MCP server returns as untrusted input, for the same reason you would treat any other external content that way.