← All terms

MCP (Model Context Protocol)

An open protocol standardizing how AI models communicate with external tools, data, and services.

What is MCP?

MCP (Model Context Protocol) is an open protocol developed by Anthropic that defines a standard way for AI models (like Claude) to communicate with external tools, databases, APIs, and other services. The goal is to create a universal "plug" between AI and the world of tools - similar to how USB-C standardized the connection of physical devices.

How MCP works

MCP introduces a server–client architecture:

• MCP server: An application or service that exposes its tools and data through a standard MCP interface (e.g., file access, database, GitHub, Slack).
• MCP client: An AI model or agent that connects to an MCP server and uses those tools when completing tasks.

Why MCP matters

• Each tool only needs to be implemented once for all compatible AI models
• Open standard - MCP servers are built by the community and new ones appear every week
• The foundation of modern AI agents that need access to real data and actions

Examples of MCP servers

• Access to file systems, databases, or Git repositories
• Integration with Slack, Google Drive, GitHub, or Notion
• Calling any REST API without writing custom code