Remote

This MCP server hosts a FastMCP-based calculator tool that exposes a single operation: add_numbers(a, b). The server is packaged in a container and is intended to run in Kubernetes (Amazon EKS) or locally via Docker. Clients connect over HTTP to the /mcp/ endpoint and interact with the MCP session using the standardized protocol: initialize a session, declare capabilities, list available tools, and then invoke add_numbers with two numeric inputs. The curl workflow in the documentation demonstrates how to initialize the session, announce readiness via notifications/initialized, list tools, and call add_numbers with arguments a and b. A Python test client is also provided to exercise the MCP server end-to-end. When you call add_numbers, the server returns a numeric result that is the sum of the two inputs.

Prerequisites:

• Docker (for building and running container locally)
• Python (optional if running outside container)
• kubectl and AWS CLI (for deploying to Amazon EKS)

Local development and testing (Docker):

1. Build the container image:

docker build -t mcp-calculator:latest .

2. Run the container locally:

docker run -p 8000:8000 mcp-calculator:latest

3. The server will be available at http://localhost:8000

If you prefer to run directly with Python (outside Docker):

1. Install Python dependencies from requirements.txt:

pip install -r requirements.txt

2. Start the server:

python server.py

3. The server will typically listen on port 8000 by default, or as configured in your environment.

Deploying to Amazon EKS (Kubernetes):

1. Ensure AWS CLI, kubectl, and Docker are installed and configured.
2. Build and push your container image to your registry (Docker Hub or ECR).
3. Update k8s-deployment.yaml to point to your image registry.
4. Deploy:

kubectl apply -f k8s-deployment.yaml

5. Check deployment status and fetch the LoadBalancer service IP:

kubectl rollout status deployment/mcp-calculator
kubectl get services mcp-calculator-service

6. Access the MCP endpoint via the LoadBalancer URL at /mcp/.

Prerequisites for EKS deployment are described in the README (AWS CLI, kubectl, Docker).

Notes and tips:

• The MCP protocol steps (especially notifications/initialized) are required after initialization.
• Sessions have a time-to-live generally around 30-60 seconds in this FastMCP setup; complete the workflow quickly or implement a keep-alive pattern if supported by your client.
• Use the provided test_mcp_client.py to validate your server locally before deploying to Kubernetes.
• If you modify the server image, update the Kubernetes deployment manifest (k8s-deployment.yaml) to reference the new image tag.
• The add_numbers tool expects inputs a and b as numbers; the response returns both a structured and plain text representation of the result.