MCP Gateway (MCPX) is a security and governance gateway that sits between AI agents and Model Context Protocol (MCP) servers, intercepting and controlling all tool invocations made by AI agents within an enterprise environment. The gateway provides a centralized point for managing access to both local and remote MCP servers through a unified registry, with built-in service discovery. It is deployable in VPCs, on-premises, or isolated networks. Core governance capabilities include: - Authentication via API keys or OAuth, with support for SSO and IAM integrations (e.g., Okta, Azure AD) - Role-based access control (RBAC) allowing fine-grained, per-agent tool restrictions with on-behalf-of attribution - Policy enforcement including rate limits, budget constraints, and approval workflows for sensitive actions - Tool customization such as overriding tool definitions and preventing tool name collisions Observability and audit features include: - Real-time telemetry capturing every MCP and tool call, including prompt payloads and tool responses - Immutable audit logs for compliance tracing - Real-time dashboards and anomaly alerts - Log streaming to external monitoring systems or SIEMs MCPX also includes an MCP Risk Score Engine (in beta) for assessing risk associated with MCP server and tool usage. The product is available as an open-source version (OSS on GitHub) and an enterprise offering. It integrates with Lunar.dev's AI Gateway for unified LLM and MCP visibility, cost tracking, and data sanitization in a single dashboard.