Abstract

The increased adoption of the Model Context Protocol (MCP) for AI Agentsnecessitates robust security for Enterprise integrations. This paper introducesthe MCP Gateway to simplify self-hosted MCP server integration. The proposedarchitecture integrates security principles, authentication, intrusiondetection, and secure tunneling, enabling secure self-hosting without exposinginfrastructure. Key contributions include a reference architecture, threatmodel mapping, simplified integration strategies, and open-sourceimplementation recommendations. This work focuses on the unique challenges ofenterprise-centric, self-hosted AI integrations, unlike existing public MCPserver solutions.

 

Quick Read (beta)