Why MCP Gateways are a Bad Idea (and What to Do Instead)   

The report argues that placing MCP Gateways as a perimeter control is the wrong approach for securing agentic systems because gateways create blind spots, introduce credential exposure, lack full agent context, become single points of failure, and are hard to enforce. It recommends implementing runtime hooks inside agent runtimes for context-aware policy enforcement and using MCP registries to allowlist approved servers and tools, providing defense-in-depth without routing secrets through third parties.