EKS上的MCP服务器
该项目使用Terraform在Amazon EKS上部署了一个模型上下文协议(MCP)服务器,用于基础设施管理。
建筑
- MCP服务器:基于Python的服务器实现MCP协议
- 容器:MCP服务器的Docker容器
- 库贝内特斯:用于容器编排的EKS集群
- 卡彭特:Kubernetes原生节点自动缩放器,实现高效缩放
- 基础设施:AWS资源管理平台
先决条件
- 配置了适当权限的AWS CLI
- Docker已安装
- 地形>=1.0
- kubectl已安装
- 已创建带有ECR存储库的AWS帐户
快速开始
- 设置环境变量:```bash
export AWS_REGION=us-east-1
export AWS_ACCOUNT_ID=your-account-id
export CLUSTER_NAME=mcp-eks-cluster
2. **使用Karpenter进行部署(推荐)**:```bash
# Option 1: Deploy everything at once
./scripts/deploy.sh all
# Option 2: Phase-by-phase deployment
./scripts/deploy.sh infrastructure # Deploy EKS cluster
./scripts/deploy.sh karpenter # Deploy Karpenter autoscaler
./scripts/deploy.sh app # Deploy MCP server
- 传统部署(不含Karpenter):```bash
make init
make plan
make apply
make deploy
## 项目结构```
├── server-enhanced.py # MCP server with real Kubernetes API integration
├── client-example.py # Example MCP client for testing
├── Dockerfile # Container definition
├── requirements.txt # Python dependencies
├── terraform/ # Infrastructure as code
│ ├── main.tf # Main Terraform configuration with integrated Karpenter
│ ├── variables.tf # Variable definitions
│ ├── outputs.tf # Output definitions
│ └── userdata.sh # EC2 user data script for Karpenter nodes
├── k8s/ # Kubernetes manifests
│ ├── namespace.yaml # Namespace definition
│ ├── deployment.yaml # Application deployment
│ ├── service.yaml # Service definition
│ └── configmap.yaml # Configuration map
├── scripts/ # Deployment scripts
│ ├── deploy.sh # Main deployment script
│ ├── terraform-init.sh # Terraform initialization
│ └── test-deployment.sh # Complete deployment test
├── claude-desktop-config.json # Claude Desktop MCP configuration
├── integrate-with-claude.md # Claude Desktop integration guide
└── Makefile # Build automation
MCP服务器功能
服务器(server-simple.py)
- Real Kubernetes API集成:使用适当的RBAC连接到实际的EKS集群
- HTTP REST API:简单的HTTP端点,便于测试和集成
- 端点:
GET /health-使用Kubernetes API连接状态进行健康检查GET /cluster-info-包含命名空间计数的实时群集信息GET /nodes-实时节点状态和信息GET /pods?namespace=-来自任何命名空间的实时pod信息GET /deployments?namespace=-部署状态和副本计数- 安全:具有最低所需权限的适当RBAC
- 错误处理:优雅的API错误处理和详细的错误响应
- 负载均衡器:通过AWS应用程序负载均衡器进行外部访问
什么是MCP(模型上下文协议)?
主控程序 使像Claude这样的人工智能助手能够连接到外部数据源和工具。除了运行kubectl命令,您还可以询问有关集群的自然语言问题!
魔术: 您的EKS集群↔️ MCP服务器↔️ 克劳德桌面↔️ 自然语言
自然语言示例
而不是像这样的命令 kubectl get nodes你可以问克劳德:
- “我的集群怎么样了?” → 获取运行状况、节点和状态
- “我的MCP服务器Pod是否正在运行?” → 检查pod运行状况并重新启动
- “我有什么Karpenter节点?” → 显示自动缩放节点
- “显示我的所有命名空间” → 列出群集命名空间
- “我的集群有问题吗?” → 全面健康检查
演示自然语言界面```bash
See what natural language queries would look like
python3 demo-natural-language.py
### 客户端集成选项
1. **HTTP REST API**:用于测试和集成的直接HTTP访问
1. **克劳德桌面**:通过MCP桥的自然语言接口(见claude desktop config.json)
1. **负载均衡器**:通过AWS ELB进行外部访问以供生产使用
## 部署命令
### 基于Karpenter的部署(推荐)```bash
# Deploy everything with Karpenter
./scripts/deploy.sh all
# Or deploy in phases for more control
./scripts/deploy.sh infrastructure # Phase 1: EKS cluster with minimal nodes
./scripts/deploy.sh karpenter # Phase 2: Karpenter installation
./scripts/deploy.sh app # Phase 3: MCP server application
# Test deployment
./scripts/test-deployment.sh
传统基础设施部署```bash
Initialize and deploy infrastructure (without Karpenter)
make init plan apply
Build and deploy application
make deploy
Complete deployment (infrastructure + app)
make full-deploy
### 局部测试```bash
# Install dependencies
pip install -r requirements.txt
# Test server locally (requires kubectl access to EKS)
python server-enhanced.py
# Run client example
python client-example.py
Claude桌面集成```bash
Follow the integration guide
cat integrate-with-claude.md
Copy configuration (macOS example)
cp claude-desktop-config.json ~/Library/Application\ Support/Claude/claude_desktop_config.json
## 配置
环境变量:
- `AWS_REGION`:AWS区域(默认:us-east-1)
- `CLUSTER_NAME`:EKS群集名称(默认值:mcp-EKS-cluster)
- `IMAGE_TAG`:Docker镜像标签(默认:最新)
- `AWS_ACCOUNT_ID`:您的AWS帐户ID
### 卡彭特配置
Karpenter设置遵循AWS EKS蓝图模式,包括:
- **EKS模块集成**:使用具有内置Karpenter支持的地形aws模块/eks
- **受管节点组**:2个带有karpenter.sh/controller污点的t3.medium节点,用于系统Pod
- **Pod标识**:使用EKS Pod Identity安全访问AWS API
- **节点池**:配置实例类型(t3.中型/大型/xlarge)和容量类型(现货/按需)
- **EC2节点类**:使用发现标签定义AMI系列(AL2023)、安全组和子网
- **自动缩放**:节点根据pod调度要求进行扩展,整合时间为30秒
- **成本优化**:更喜欢定点实例并整合未充分利用的节点
### 当前状态✅
- **卡彭特**:成功部署和配置节点(AL2023 AMI)
- **MCP服务器**:运行在Karpenter-托管节点上的HTTP REST API
- **负载均衡器**:外部访问已配置并正常工作
- **RBAC**:Kubernetes API访问的正确权限
## 监控
访问您的MCP服务器:```bash
kubectl get service mcp-server-service -n mcp-server
查看日志:```bash
kubectl logs -f deployment/mcp-server -n mcp-server
## 使用示例
### HTTP REST API
通过HTTP端点直接访问服务器:```bash
# Get LoadBalancer URL
LB_URL=$(kubectl get service mcp-server-service -n mcp-server -o jsonpath='{.status.loadBalancer.ingress[0].hostname}')
# Health check
curl http://$LB_URL/health
# Cluster information
curl http://$LB_URL/cluster-info
# Node information
curl http://$LB_URL/nodes
# Pod information
curl http://$LB_URL/pods?namespace=mcp-server
# Deployment information
curl http://$LB_URL/deployments?namespace=default
局部测试```bash
Port forward for local access
kubectl port-forward -n mcp-server service/mcp-server-service 8080:80
Test endpoints
curl http://localhost:8080/health
curl http://localhost:8080/cluster-info
## 测试
### 快速测试
运行综合测试套件:```bash
./test-all.sh
手动测试```bash
Get LoadBalancer URL
LB_URL=$(kubectl get service mcp-server-service -n mcp-server -o jsonpath='{.status.loadBalancer.ingress[0].hostname}')
Test all endpoints
curl http://$LB_URL/health
curl http://$LB_URL/cluster-info
curl http://$LB_URL/nodes
curl http://$LB_URL/pods?namespace=mcp-server
curl http://$LB_URL/deployments?namespace=default
### 测试卡彭特自动缩放```bash
# Apply test workload to trigger node provisioning
kubectl apply -f test-karpenter.yaml
# Watch Karpenter create new nodes
kubectl get nodeclaim -w
# Scale up to test more provisioning
kubectl scale deployment test-workload --replicas=8
# Check new nodes
kubectl get nodes -l karpenter.sh/nodepool
# Clean up
kubectl delete -f test-karpenter.yaml
监视器Karpenter```bash
Check Karpenter status
kubectl get nodepool,ec2nodeclass -o wide
Watch Karpenter logs
kubectl logs -f -n karpenter -l app.kubernetes.io/name=karpenter
Check node provisioning
kubectl get nodeclaim
## 故障排除
### 常见问题
1. **导入错误**:安装依赖项 `pip install -r requirements.txt`
1. **Kubernetes API访问**:确保 `kubectl` 已配置并正在工作
1. **ECR权限**:验证AWS凭据是否具有ECR访问权限
1. **Claude桌面连接**:检查配置文件路径和格式
### 日志和调试```bash
# Check pod logs
kubectl logs -f deployment/mcp-server -n mcp-server
# Test Kubernetes connectivity
kubectl get nodes
# Verify ECR repository
aws ecr describe-repositories --repository-names mcp-server
清理
Karpenter部署清理```bash
Remove in reverse order
kubectl delete -f k8s/
cd terraform && terraform destroy -auto-approve
### 遗留部署清理```bash
make destroy
