Model Context Protocol (MCP)

The Model Context Protocol (MCP) is a communication standard that allows large language models (LLMs) to connect with external tools and services in a consistent and reliable way.

On their own, LLMs are limited to generating text, images, or other outputs. MCP expands their capabilities by acting as a bridge, letting models interact with resources like databases, cloud apps, email services, and APIs without having to create custom integrations for each one.

These are the four components of MCP.

• Host: The environment where the LLM operates, such as a chatbot or AI agent. The host is what users interact with directly and routes requests between the model and external resources.
• MCP client: Lives inside the host and acts as a bridge between the LLM and MCP servers. It handles requests from the model, translates them into a protocol-compliant format, sends them to the correct server, and returns responses to the host.
• MCP server: This is managed by tool providers and exposes tools, prompts, and resources in a standardized way.
• Transport Layer: Handles how messages move between clients and servers.

MCP is one of many protocols available to AI systems. Others include Google’s Universal Commerce Protocol (UCP) and Agent2Agent Protocol (A2A).