MCP Client
An MCP client is the component inside an MCP host that connects to one MCP server, sends the agent's tool calls to it and handles the responses. A host such as Cursor or Claude Desktop runs a separate client for each server it uses, which is why one host is several MCP clients at once.
Why It Matters
Claude Desktop, Cursor and Visual Studio Code all get called MCP clients. The MCP specification splits the term in two. An MCP host is the application a person works in. An MCP client is the component inside that host which handles the connection to a single MCP server. Visual Studio Code connected to a Sentry server and a filesystem server is one MCP host running two MCP clients.
Clients implement different parts of MCP, which is why the same MCP server can behave differently in two applications. When a server needs an answer from the person before it can finish, a client that supports elicitation shows the question and a client that does not shows nothing. Enough of these gaps exist that people publish compatibility tables comparing the widely used clients.
One program can be an MCP client and an MCP server at the same time. An MCP gateway works this way. It answers the agent in front of it as a server, and connects to each MCP server behind it as a client, which is how a single endpoint can present the tools of many servers. Anyone reading logs from that setup is looking at two hops rather than one.