MCP Gateways sit between Model Context Protocol clients and backend MCP servers to provide enterprises with centralized governance, security, and operational control over AI agent tool use. They terminate inbound MCP connections from LLM applications, route requests to appropriate upstream servers or tool backends, and enforce control-plane policies including authentication, authorization, audit logging, and runtime guardrails. Core capabilities span identity federation with enterprise SSO, fine-grained permission models that restrict which tools specific users or applications may invoke, request and response filtering to prevent data leakage, and observability for compliance reporting. The category emerged in late 2024 alongside the Model Context Protocol specification itself, initially as open-source proxies and rapidly evolving into managed platforms that integrate policy engines, secret management, and multi-tenant isolation as organizations move agentic AI workloads into production.