MCP Gateway — Route and Manage MCP Servers

Practical Notes

• Quant: start with 1 sample MCP server image and measure rollout time end-to-end.
• Quant: define SLOs (p95 latency, error rate) for gateway vs direct stdio usage.

Pattern: separate tool hosting from tool use

A gateway becomes useful when you have multiple MCP servers and multiple clients.

Operational guidance:

• Treat MCP servers as deployable artifacts (images) with versioning.
• Use the gateway as the control plane for discovery and routing.
• Keep auth/network policies at the gateway boundary.

Migration tip

Start by routing a single non-critical tool through the gateway, then expand coverage once observability and failure handling are solid.

FAQ

Q: Is this only for Kubernetes? A: The README focuses on Docker + Kubernetes deployments, plus an Azure deployment path.

Q: What should I gateway first? A: A low-risk tool with clear success criteria and measurable latency.

Q: How do I keep versions sane? A: Version images, pin gateway releases, and roll out with canaries + metrics.