MCP Server for Kubernetes
A server for managing Kubernetes clusters using the Model Context Protocol.
MCP Kubernetes Server
A Model Context Protocol (MCP) server for Kubernetes, supporting chunked HTTP streaming, advanced troubleshooting prompts, and full Kubernetes resource/tool coverage.
Prerequisites
- Node.js v18+ (for local dev/build)
- Docker (for containerization)
- Access to a Kubernetes cluster (AKS, EKS, GKE, or local)
kubectlconfigured (for testing and kubeconfig management)
1. Build and Run Locally
npm install
npm run build
MCP_TRANSPORT=http-chunked npm start
- By default, the server uses your local kubeconfig (
~/.kube/configorC:\Users\<username>\.kube\config). - To use a custom kubeconfig, set the
KUBECONFIGenvironment variable:export KUBECONFIG=/path/to/your/kubeconfig npm start
2. Dockerize the MCP Server
Build the Docker image
docker build -t yourrepo/mcp-server:latest .
Push to your registry
docker push yourrepo/mcp-server:latest
3. Deploy on Kubernetes (AKS, EKS, GKE)
Edit the image name in k8s-mcp-server.yaml:
Replace yourrepo/mcp-server:latest with your image name.
Apply the manifest
kubectl apply -f k8s-mcp-server.yaml
- This creates a namespace, ServiceAccount, RBAC, Deployment, and Service.
- By default, the Service is
ClusterIP(internal). Change toLoadBalancerorNodePortfor external access.
4. Using the MCP Server
HTTP Chunked Endpoint
- The server exposes
/call-tool-chunkedon port 3000. - Example (using
curl):curl -X POST http://<server-ip>:3000/call-tool-chunked \ -H "Content-Type: application/json" \ -d '{"name": "get_pods", "args": {"namespace": "default"}}' - The response will stream progress and results as JSON lines.
Using Prompts
- To use a prompt, POST to
/call-tool-chunkedwith the prompt name, e.g.:curl -X POST http://<server-ip>:3000/call-tool-chunked \ -H "Content-Type: application/json" \ -d '{"name": "k8s-pod-crashloop-diagnose", "args": {"podName": "my-pod", "namespace": "default"}}'
5. Kubeconfig and Permissions
- The MCP server uses the kubeconfig available in the container (default:
/root/.kube/config). - For in-cluster deployments, it uses the ServiceAccount and RBAC provided in the manifest.
- To use a custom kubeconfig, mount it as a secret and update the Deployment (see commented lines in the manifest).
6. Security Notes
- Do not expose the MCP server to the public internet without authentication and TLS.
- Use network policies, firewalls, or VPNs to restrict access.
- Use least-privilege RBAC for the ServiceAccount.
7. Extending and Customizing
- Add new tools, resources, or prompts in the
src/directory. - Rebuild and redeploy the Docker image after making changes.
8. Troubleshooting
- Check logs with
kubectl logs -n mcp-server deploy/mcp-server. - Ensure the ServiceAccount has the required permissions for your use case.
- For local testing, ensure your kubeconfig is valid and has cluster access.
License
MIT
संबंधित सर्वर
Coinbase Chat MCP
Access cryptocurrency data, market information, and trading insights from Coinbase's public API.
Umbrix MCP Server
An MCP server for the Umbrix CTI platform, providing AI-powered threat intelligence.
Vertex AI MCP Server
Interact with Google Cloud's Vertex AI Gemini models for coding assistance and general query answering.
FreshMCP
Provides an MCP interface for FreshMCP operations using Azure Cosmos DB and AI Search.
Merlin Energy — BESS Quoting & Sales Intelligence
AI-powered BESS quoting & energy sales agent for Claude and other MCP clients. Generate TrueQuote™ estimates, qualify leads, compare competitors, and produce proposals — in seconds.
Lodgify MCP Server
An MCP server for the Lodgify vacation rental API.
Beyond Menu Salesforce MCP
Integrate with Salesforce to perform actions like testing connections and running queries.
Preloop
Preloop is an event-driven AI automation platform with built-in human-in-the-loop safety based on MCP proxying.
Datadog MCP Server
Provides comprehensive Datadog monitoring capabilities through any MCP client.
Cloudflare Remote MCP Server (Authless)
Deploy a remote, authentication-free MCP server on Cloudflare Workers or locally via npm.