The Model Context Protocol is an open standard, originally developed by Anthropic and now adopted by other AI providers, that defines how AI applications (the "host" — typically a chat client or an agent runtime) connect to external tools, data sources, and APIs (the "servers"). Where ad-hoc integrations between LLMs and external systems are typically bespoke and opaque, MCP makes them declarative, inspectable, and consistent across providers.
For data governance, MCP matters because it makes the boundary between a model and its data sources explicit. Each MCP server declares the resources it exposes, the tools it offers, and the authentication it requires. The host can enumerate, log, and gate every interaction. That auditability is a precondition for treating AI systems as part of a governed data architecture rather than a black box that quietly reaches into everything.
The standard is young — adopted broadly in 2024–2025 — but the architectural pattern is foundational. Expect AI governance frameworks to start prescribing MCP-style declarative interfaces between AI agents and the systems they touch, much the way information security shifted from ad-hoc integrations to standardized OAuth scopes and service accounts over the last decade.