What is MCP?
How the Model Context Protocol gives agents a standard way to discover and call tools from external systems.
A common interface for agent tools
The Model Context Protocol (MCP) is an open protocol for connecting AI applications to tools and context providers. An MCP server publishes capabilities; an MCP client discovers them and makes them available to a model.
Instead of writing a custom integration contract for every agent framework, teams can expose a tool once and connect compatible clients to it.
How a call works
A client connects to a server, requests its tool catalog, and receives names plus JSON schemas for their arguments. When the model selects a tool, the client validates and sends the call to the server.
- Discovery describes what is available.
- Schemas constrain the arguments a model can send.
- Transport carries requests over stdio or a network connection.
- The host remains responsible for consent, credentials, and policy.
MCP is not authorization
MCP standardizes connectivity; it does not decide whether a particular agent should be allowed to issue a refund or read a customer record. Apply least-privilege permissions and runtime policies around MCP tool calls.