What is an MCP client?

An MCP client is software that runs the Model Context Protocol, discovers a server's capabilities, and invokes tools, resources, and prompts on the server's behalf. A client is driven by a host — the application with the user or model context. Clients connect to servers, transmit protocol requests, and deliver responses back to the host.

The MCP specification separates the host from the client. The host is the application that holds context — the desktop app, the coding tool, or the agent runtime. The client is the protocol-facing component it uses. Separating the two lets one host manage many server connections.

A client lifecycle mirrors the MCP initialization sequence. The client sends an initialize request, receives the server's capabilities, exchanges an initialized notification, then performs primitive operations: tools/list, resources/list, prompts/list, then the corresponding call/read/get requests.

A client calls a tool with tools/call, supplying arguments that match the tool's declared schema, and receives content back. For resources, a client reads them by URI. Which primitives are available is negotiated during initialization.