Skip to content

MCP client ​

The part of an AI application that holds one connection to one MCP server: it negotiates the protocol version and capabilities, then sends that server's requests and receives its replies.

In MCP, the "client" is not the app you install. The app is the hostMCP host: The AI application, such as a chat app, an IDE or an agent runtime, that creates MCP clients, connects them to servers and decides what the model may use., and the host creates one client for each server it connects to. The specification puts it plainly: each client has "a 1:1 relationship with a particular server". An IDE connected to a file server, a ticket tracker and a search API runs three clients side by side.

A client's job is narrow but strict:

  • Open the session. It sends initialize first, with the latest protocol version it supports and its own capabilities (for example, whether it can answer sampling or elicitation requests). If the server answers with a version the client does not support, the client should disconnect.
  • Respect what was agreed. Both sides may only use the capabilities they declared.
  • Keep servers apart. One server's client never sees another server's traffic. Combining results is the host's job, which is why the host, not the client, decides what the model may call.

Example ​

A host connects to a remote ticket server. Its client posts:

json
{ "jsonrpc": "2.0", "id": 1, "method": "initialize",
  "params": { "protocolVersion": "2025-11-25", "capabilities": {},
              "clientInfo": { "name": "my-agent", "version": "1.0.0" } } }

The server replies with 2025-11-25 and "tools": { "listChanged": true }. The client sends notifications/initialized, then tools/list, and hands the tool list to the host.

Common confusion ​

People often call the whole chat app or IDE "the MCP client". That is fine as shorthand, but in the spec the host owns consent, user authorization and the model; the client is just one connection. When a security rule says "the client MUST", check what it is about: protocol rules (headers, versions, token audience) fall on the connection; rules about consent and what people see fall on the host.

In L2M ​

L2M is an MCP client whenever an agent or MCP Tool node calls a server: it connects over Streamable HTTPStreamable HTTP: The MCP transport for servers reached over a network: every client message is an HTTP POST to one endpoint, and the server answers with a JSON body or a server-sent events stream. or stdiostdio transport: The MCP transport in which the client starts the server as a child process and exchanges newline-delimited JSON-RPC messages over the process's standard input and output. and, by default, asks for protocol version 2025-06-18 in initialize and, over HTTP, sends the negotiated MCP-Protocol-Version header afterwards. If a server speaks only 2026-07-28 and rejects initialize, L2M switches to 2026-07-28 for that server. Trying 2026-07-28 first is opt-in, per connection (protocolMode) or deployment-wide (MCP_CLIENT_PROTOCOL_DEFAULT). See the MCP server registry. Through the MCP gateway, L2M also plays the server role for other clients.

Further reading ​

This website (not the L2M product) uses Google Analytics, only after you accept cookies. Cookie settings