mcp-server

By: Zack Chaffee A20478873

A server implementing Model Coupling Protocol (MCP) capabilities for HDF5 file operations and Slurm job management.

Features

HDF5 file operations:

Read datasets
List file contents

Slurm job management:

Submit jobs
Check job status

Node Hardware Operations

Get CPU information
Get memory information
Get disk information
Get comprehensive system information

Compression Operations

Compress string data with gzip or zlib
Compress files with gzip or zlib
Decompress data

Initialization

Once you clone this reponsitory cd into it

After this hwe will create a virtual enviornment and install all dependincies:

uv venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
uv pip install -e .
uv pip install -e ".[test]"

Running

To startup the server run:

python -m src.server

This will autoclocate the server at http://localhost:8000.

Endpoints

POST /mcp: Main endpoint for MCP requests
GET /health: Health check endpoint

Examples:

import httpx

async with httpx.AsyncClient() as client:
    # Read a dataset
    response = await client.post("http://localhost:8000/mcp", json={
        "capability": "hdf5",
        "action": "read_dataset",
        "parameters": {
            "file_path": "/path/to/data.h5",
            "dataset_path": "/path/to/dataset"
        }
    })
    
    # List contents
    response = await client.post("http://localhost:8000/mcp", json={
        "capability": "hdf5",
        "action": "list_contents",
        "parameters": {
            "file_path": "/path/to/data.h5",
            "group_path": "/"
        }
    })

curl -X POST http://localhost:8000/mcp \
  -H "Content-Type: application/json" \
  -d '{
    "jsonrpc": "2.0",
    "method": "mcp/listTools", 
    "params": {}, 
    "id": "1"
  }'

Testing

For testing rung:

pytest

For tests with coverage:

pytest --cov=src