A Python-based implementation of a Model Context Protocol (MCP) server that provides weather information using the OpenWeather API as a tool.
This project demonstrates how to build and deploy an MCP-compliant server that uses the FastMCP SDK to expose weather-related tools on an Amazon EKS cluster. The server supports both streaming (SSE) and non-streaming modes, allowing clients to retrieve current weather data and forecasts for cities around the world through an Application Load Balancer (ALB).
- Amazon EKS: Managed Kubernetes service for running containerized applications
- BottleRocket MNG: Secure, minimal-footprint Linux-based OS for container hosting
- AWS ALB Controller: Manages AWS Application Load Balancers for Kubernetes ingress
- AWS ECR: Container registry for storing Docker images
- FastAPI Application: The core server built with FastAPI, providing HTTP endpoints that comply with the MCP specification
- FastMCP Integration: Uses the MCP Python SDK to register and expose weather tools
- OpenWeather API Client: Retrieves real-time weather data and forecasts
- SSE Transport: Implements Server-Sent Events for streaming responses
- MCP Client: A Python client that demonstrates how to interact with the MCP server
This project demonstrates how to deploy an MCP server on Amazon EKS with the following features:
- BottleRocket Managed Node Groups: Secure, purpose-built OS for running containers
- ALB Ingress Controller: Automatically provisions AWS Application Load Balancers
- ECR Integration: Stores and manages container images
- Kubernetes Secrets: Securely manages the OpenWeather API key
The server implements the Server-Sent Events (SSE) protocol for streaming responses, which offers several advantages:
- Long-lived Connections: Allows the server to push updates to the client over a single HTTP connection
- Simple Protocol: Uses standard HTTP with a specialized content type (
text/event-stream) - Automatic Reconnection: Browsers automatically attempt to reconnect if the connection is dropped
- Event Formatting: Messages are sent with a
data:prefix and separated by double newlines
The server registers two main weather tools using the FastMCP SDK:
- get_current_weather: Retrieves and formats current weather conditions for a city
- get_weather_forecast: Provides a 5-day weather forecast with daily temperature ranges and conditions
Tools are registered in the FastMCP instance's tools dictionary, making them accessible through:
result = await weather_mcp.tools["get_current_weather"].function(city=city)- Python 3.10 or higher
- Docker
- kubectl configured for your EKS cluster
- AWS CLI configured with appropriate permissions
- Terraform (for infrastructure deployment)
-
Clone the repository:
git clone <repository-url> cd mcp-eks-demo -
Put your openweather API_Key in .tfvar file
openweather_api_key=<your_api_key> -
Deploy the EKS cluster with the command
terraform apply --var-file=terraform.tfvars --auto-approve
This command will create the EKS Cluster with BottleRocket AMI based Managed Node Groups and also create kubernetes secret from the openweather API Key.
- Build and deploy the MCP server:
cd mcp-eks-demo/mcp ./build-deploy.sh
This will build the Docker container with the MCP Server and the tools code, create an ECR repository and push the image. It will also create deployment YAML with Deployment, Service and Ingress resources.
- Get the ALB URL:
export ALB_URL=$(kubectl get ingress mcp-server-py-ingress -n default -o jsonpath='{.status.loadBalancer.ingress[0].hostname}')
The MCP client supports both current weather and forecast queries:
-
For current weather:
python mcp_client.py $ALB_URL "London" -
For weather forecasts:
python mcp_client.py $ALB_URL "forecast for New York"
OPENWEATHER_API_KEY: Required for API accessPORT: Server port (default: 8000)LOG_LEVEL: Logging level (default: INFO)HTTPX_TIMEOUT: HTTP request timeout in seconds (default: 5)
- API Key Management: The OpenWeather API key is stored as a Kubernetes secret
- Network Security: Traffic is managed through the ALB, which can be configured with appropriate security groups
- Container Security: BottleRocket provides enhanced security for container workloads with minimal attack surface
The server implements comprehensive error handling:
- API timeouts
- City not found errors
- General API errors
- Session management errors