TL;DR: The Model Context Protocol is an open standard that enables language models to interface with external tools and information sources through a unified JSON-RPC 2.0 mechanism.

How MCP Works

MCP operates through JSON-RPC 2.0 across a connection between three components:

Hosts - The language model application users interact with (Claude Desktop, Cursor, ChatGPT)

Clients - The connector within the host communicating with servers

Servers - Lightweight services exposing specific capabilities (GitHub, Slack, PostgreSQL, filesystem)

Each server provides three core capabilities:

Resources - Read-only contextual information

Prompts - Reusable templated prompts users can select

Tools - Functions models can invoke

Why MCP Exists: The M-by-N Problem

Previously, each language model application required custom integration with third-party tools. With M models and N tools, this created M x N necessary integrations. MCP reduces this to M+N. By early 2026, over 500 public MCP servers existed.…