kube-MCP
A Model Context Protocol server for Kubernetes that provides full resource coverage and advanced troubleshooting tools via HTTP chunked streaming. It enables users to manage clusters and diagnose complex issues like pod crashloops through specialized prompts and standard kubectl-like operations.
README
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
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。