MCP (Model Context Protocol)
Model Context Protocol (MCP) is an open standard for connecting AI applications to external tools, data and prompt templates. An MCP server wraps one system, such as a database, ticket tracker or file store, and any MCP client can use it. One connector then works across many assistants and AI agents.
How it works
MCP has three roles. The host is the AI application, a client inside the host holds one connection to each server, and a server exposes tools, resources and prompts. Messages use JSON-RPC 2.0, over stdio for a local server or Streamable HTTP for a remote one. A client discovers tools with tools/list and runs one with tools/call. Without a standard, 3 assistants and 5 internal systems need 3 × 5 = 15 custom integrations. With MCP you build 5 servers and 3 clients, 8 pieces in total.
Why it matters in production
An MCP server hands a model real permissions, so it is part of your attack surface. Scope each server to the least access it needs, keep write tools behind human approval, and log every tool call. Treat tool descriptions and returned data as untrusted, because they can carry prompt injection.
See also
On the One2N blog
Further reading
