Enterprises have long secured cloud servers and corporate laptops with Mobile Device Management. But the rise of AI workloads has created a blind spot: local AI traffic from tools like Claude Desktop. This traffic often bypasses security and observability, creating a gap in defense-in-depth strategies.

The solution involves treating AI traffic like any other enterprise data. By routing Claude Desktop calls through an agentgateway on Kubernetes, organizations can gain visibility and control. This setup intercepts traffic between the user and MCP Servers, applying security rules and logging at the gateway level.

Deploying this requires a Kubernetes cluster, Claude Desktop, and agentgateway. The process starts with deploying an MCP Server (like a math service), configuring the gateway to route to it, and finally pointing Claude's configuration to the gateway's endpoint. This ensures all AI tool interactions are centralized and observable.