MCP Is the Tool Plane, Not the Agent Controller: Enterprise Agent Control Plane Series, Part 1

This article argues that the Model Context Protocol (MCP) should serve as the governed interface for tool discovery and invocation in enterprise AI, but not as the full controller for autonomous agents. It emphasizes a clean separation: the agent proposes actions, the controller authorizes them, MCP carries the interaction, and the runtime executes and verifies. The author warns against overloading MCP with workflow, policy, approval, and lifecycle management responsibilities, which should belong to a separate Agent Control Plane. Part 1 of a series, it sets the stage for designing trusted agent controllers with distinct architectural concerns like identity, permissions, risk, and production safety.