MCP Server Implementation

Name: Esteban Nicolas Student ID: A20593170

I. Implemented MCP Capabilities

1 Data Resources 1.1 HDF5 File Listing

• Lists mock HDF5 files in a directory structure
• Parameters: path_pattern (optional file path pattern)

2 Tools 2.1 Slurm Job Submission

• Simulates job submission to a Slurm scheduler
• Parameters: script_path (required), cores (optional, default=1)

2.2 CPU Core Reporting

• Reports number of CPU cores available on the system
• No parameters required

2.3 CSV Visualization

• Plots two columns from a CSV file (defaults to first two columns)
• Parameters: csv_path (required), column x, column y (both optional)

II. Setup Instructions

1. Create virtual environment

uv venv -p python3.10 .venvScriptsactivate # On Unix: source .venv/bin/activate

2. Install dependencies

uv sync uv lock

3. Environment configuration The project uses pyproject.toml for dependency management. Key dependencies include:

FastAPI

Uvicorn

Pydantic

Pandas

Matplotlib

Pytest

Pytest-ascyncio

4. Running the MCP Server

Start the server cd src uvicorn server:app --reload

The server will be available at:

API endpoint: http://localhost:8000/mcp Health check: http://localhost:8000/health

III Testing

1. Run all tests:

pytest tests/ Run specific test file:

pytest tests/test_capabilities_plot_vis.py pytest tests/test_capabilities_hdf5.py pytest tests/test_capabilities_cpu_core.py pytest tests/test_capabilities_slurm.py pytest tests/test_mcp_handler.py

2. Example Requests 2.1 List available resources

curl -X POST http://localhost:8000/mcp
-H “Content-Type: application/json”
-d ‘{“jsonrpc”:“2.0”,“method”:“mcp/listResources”,“id”:1}’

2.2 List HDF5 files

curl -X POST http://localhost:8000/mcp
-H “Content-Type: application/json”
-d ‘{“jsonrpc”:“2.0”,“method”:“mcp/callTool”,“params”:{“tool”:“hdf5_file_listing”,“path_pattern”:“/data/sim_run_123”},“id”:2}’

2.3 Submit Slurm job

curl -X POST http://localhost:8000/mcp
-H “Content-Type: application/json”
-d ‘{“jsonrpc”:“2.0”,“method”:“mcp/callTool”,“params”:{“tool”:“slurm_job_submission”,“script_path”:“/jobs/analysis.sh”,“cores”:4},“id”:3}’

2.4 Plot CSV columns

curl -X POST http://localhost:8000/mcp
-H “Content-Type: application/json”
-d ‘{“jsonrpc”:“2.0”,“method”:“mcp/callTool”,“params”:{“tool”:“plot_vis_columns”,“csv_path”:“data.csv”,“column x”:“time”,“column y”:“temperature”},“id”:4}’

IV Implementation Notes

1. Mock Implementations:

-HDF5 file listing uses a simulated directory structure -Slurm job submission generates mock job IDs -CPU core reporting uses os.cpu_count()

2. CSV Visualization:

-Creates plots in a plots_results directory -Defaults to first two columns if none specified -Returns path to generated PNG file

3. Error Handling:

-Proper JSON-RPC 2.0 error responses -Input validation for all parameters -Graceful handling of missing files/invalid paths

GITHUB: https://github.com/EstebanIIT/cs550_MCP.git