Simon Willison argues that the Model Context Protocol holds practical value for systems requiring stricter security controls, even as full-featured terminal agents like Claude Code operate without it. He notes that unrestricted internet access is not always desirable or safe for every deployment scenario. Instead, organisations often need granular control over which external services an agent can reach. They also require authentication methods that prevent direct exposure of API keys. A user-friendly interface for connecting and authenticating further services becomes necessary in these constrained environments. Finally, strong audit logging is essential to track exactly what actions an automated system performs. The protocol simplifies the implementation of these four specific requirements.
This matters because dismissing the protocol as obsolete ignores the needs of non-YOLO applications where safety and oversight are prioritised over raw speed. It ensures that sensitive infrastructure remains protected while still allowing automation to function within defined boundaries. The approach supports a wider range of enterprise use cases beyond simple code generation.
* Direct API key exposure is prevented through indirect authentication methods
* User interfaces handle the complexity of connecting external services securely
* Audit logs provide a clear record of all agent interactions
