A new InfoQ article argues that production deployments of the Model Context Protocol need security controls beyond a single gateway layer.
The piece focuses on MCP, a protocol that lets AI applications connect to external tools and data sources. It proposes four control layers: safe execution, management infrastructure, outbound trust, and semantic integrity. The central claim is that enforcement should happen at the earliest reliable control point, not only after traffic reaches a gateway.
That matters because MCP can give models access to sensitive systems. A gateway may help centralize policy, but it cannot fully solve problems such as untrusted tool output, unsafe execution environments, or requests that look valid while carrying harmful meaning.
For teams adopting agentic workflows, the article is a reminder that tool access is security architecture. MCP deployments need isolation, identity, monitoring, and content-aware checks if they are going to move safely from experiments into production.