Model Context Protocol (MCP) is an open standard that defines how machine intelligence connects to external tools, data sources, and services through a single protocol — one adapter instead of a custom wire for every pair.

Before MCP, integrating a model with a tool meant bespoke code on both sides. Add five tools and three models and you have fifteen integration points to maintain. MCP collapses that to N-plus-M: tool providers implement one server, model providers implement one client, and any combination works. For teams building AI agents or systems that need to act on real infrastructure, this is the difference between a composable stack and a pile of duct tape.

Anthropic open-sourced MCP in November 2024. Adoption was fast — OpenAI, Google, and others followed, and governance moved to the Linux Foundation's Agentic AI Foundation. That last move matters: it signals the protocol is shared infrastructure, not a proprietary moat. Betting your integration layer on MCP is now a reasonable call.

The practical upside is portability. An MCP server you build today works with whatever model is best next year. That's the real value — not the protocol itself, but the option value it preserves.