When is Kubernetes worth it?

When you need multi-service operations, scaling, policy, and existing platform expertise.

Can MCP scale horizontally?

Stateless HTTP servers can scale horizontally if sessions and streams are handled correctly.

Should each server get a namespace?

Sensitive environments often benefit from namespace separation and scoped secrets.

Should I scale on CPU or memory?

Both. Tools that buffer large outputs or hold open streaming connections can become memory-bound before CPU-bound, so a memory target catches scaling needs a CPU-only HPA would miss.

How should secrets reach the pod?

Prefer an external secrets operator syncing from a manager such as AWS Secrets Manager or Vault over hand-maintained Kubernetes Secret manifests, so rotation does not require a manual redeploy.

deploymentHowTo

Deploy MCP on Kubernetes

Run MCP servers on Kubernetes with deployments, secrets, ingress, scaling, probes, and observability.

Quick Answer / TL;DR

Kubernetes fits MCP platforms that need multiple services, strict isolation, custom networking, autoscaling, secret management, and mature observability.

Key Takeaways

Use probes.
Separate namespaces by environment.
Centralize logs and policies.
Externalize secrets rather than storing them in plain Kubernetes Secret manifests.

Minimal deployment

Kubernetes is powerful but operationally heavier. Use it when the organization already runs clusters or needs strong multi-service orchestration.

yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: payments-mcp
spec:
  replicas: 2
  selector:
    matchLabels:
      app: payments-mcp
  template:
    metadata:
      labels:
        app: payments-mcp
    spec:
      containers:
        - name: server
          image: registry.example.in/payments-mcp:1.0.0
          ports:
            - containerPort: 8080

Multi-stage Docker build

Build with a multi-stage Dockerfile so the production image ships only compiled output and production dependencies, not the TypeScript toolchain, and run the process as a non-root user.

text

FROM node:20-alpine AS builder
WORKDIR /app
COPY package*.json ./
RUN npm ci --only=production
COPY . .
RUN npm run build

FROM node:20-alpine
WORKDIR /app
COPY --from=builder /app/node_modules ./node_modules
COPY --from=builder /app/dist ./dist
USER node
EXPOSE 3000
CMD ["node", "dist/index.js"]

Autoscaling and externalized secrets

Scale on both CPU and memory utilization rather than CPU alone, since MCP tool handlers that buffer large tool outputs can be memory-bound before they are CPU-bound. Pull secrets from a manager such as AWS Secrets Manager or Vault instead of committing them as plain Kubernetes Secret manifests.

yaml

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: mcp-server-hpa
spec:
  scaleTargetRef: { apiVersion: apps/v1, kind: Deployment, name: mcp-server }
  minReplicas: 2
  maxReplicas: 10
  metrics:
    - type: Resource
      resource: { name: cpu, target: { type: Utilization, averageUtilization: 70 } }
    - type: Resource
      resource: { name: memory, target: { type: Utilization, averageUtilization: 80 } }