Polp
← Back to glossary
AI agents

Model Context Protocol (MCP)

An open protocol for connecting AI applications to tools, data sources, and reusable context through a consistent interface.

Business example

An approved MCP server lets an assistant query a calendar or internal system without a bespoke integration for every model.

How it works

  1. 1.A server exposes resources and tools.
  2. 2.A client discovers their schemas and capabilities.
  3. 3.The host applies identity, permissions, and user approval.

Common misconceptions

MCP makes every connected tool safe.
The protocol standardizes access; the host must still enforce authentication, authorization, and review.

Related concepts

AI agentPrompt injection

Related reading

MCP vs A2A: the difference every company should understand

Sources

Reviewed:

Reviewed by: Javier Chulvi Bernad · LLM Engineer · Madrid