Summary

Implement Kubernetes Service runtime to connect to MCP servers deployed as long-running K8s Services, ideal for shared, always-available tools.

Parent Epic

Part of #1 - Production Kubernetes & Container Support

Use Cases

• Connecting to existing MCP servers deployed in the cluster
• Shared tools across multiple gateway instances
• Long-running services with high availability
• Database connections, monitoring, external APIs

RuntimeConfig Addition

pub enum RuntimeConfig {
    // ... existing variants
    KubernetesService {
        namespace: String,
        service_name: String,
        port: u16,
        protocol: TransportProtocol,  // SSE or Stdio-over-HTTP
        path: Option<String>,
        tls: Option<TlsConfig>,
    },
}

Catalog Example

servers:
  - id: shared-postgres
    runtime:
      type: k8s-service
      namespace: mcp-servers
      service_name: postgres-mcp
      port: 8080
      protocol: sse
      path: /mcp

  - id: internal-api
    runtime:
      type: k8s-service
      namespace: platform
      service_name: internal-api-mcp
      port: 443
      protocol: sse
      tls:
        ca_cert: /var/run/secrets/ca.crt

Implementation Details

Service Discovery

1. Resolve service DNS: {service_name}.{namespace}.svc.cluster.local
2. Connect via cluster networking
3. Establish SSE or HTTP connection
4. Health check via K8s endpoints API

Health Monitoring

• Watch Endpoints for service health
• Automatic reconnection on pod restarts
• Circuit breaker for failing services

Features Required

• Service DNS resolution within cluster
• SSE transport over HTTP/HTTPS
• TLS certificate configuration
• Health check integration
• Connection pooling
• Automatic reconnection

Acceptance Criteria

• Can define server with type: k8s-service in catalog
• Connects to service within the cluster
• Works with ClusterIP services
• TLS/mTLS support
• Handles service endpoint changes gracefully
• Health status reflects service availability

Dependencies

• Depends on feat: Implement k8s-job runtime for ephemeral workloads #5 (kube-rs client integration)
• Requires RBAC: services/get, endpoints/watch

References

• kube-rs
• Kubernetes Services