Definition
An open standard that allows AI agents and models to safely connect to external tools, data sources, and APIs without custom integration code. MCP defines how an agent can request tool capabilities, what data flows between agent and tool, and what permissions apply.
Why it matters
MCP has become the de facto standard for agentic integrations across Claude, Cursor, GitHub Copilot, and dozens of enterprise AI platforms. It is the primary attack surface for tool-based AI compromise; broken MCP implementations create pathways for privilege escalation and data exfiltration.