mcp-server

What is mcp-server?

mcp-server is a Python server that implements the Model Coupling Protocol (MCP) for HDF5 file operations, Slurm job management, node hardware introspection, and data compression. It is designed for HPC or scientific computing environments where MCP workflows need to interact with HDF5 datasets, submit Slurm jobs, query system hardware, or compress/decompress data.

How to use mcp-server?

Clone the repository, create a virtual environment with uv venv .venv, activate it, and install dependencies with uv pip install -e . (optionally -e ".[test]"). Start the server by running python -m src.server, which listens at http://localhost:8000. Send MCP requests to the POST /mcp endpoint or use a health check at GET /health.

Key features of mcp-server

• Read HDF5 datasets and list file contents
• Submit Slurm jobs and check job status
• Get CPU, memory, disk, and comprehensive system information
• Compress/decompress string data or files using gzip or zlib
• JSON-RPC interface via HTTP for tool discovery and invocation

Use cases of mcp-server

• Integrate HDF5 data access into MCP-based AI agents
• Manage Slurm job lifecycles from within an MCP workflow
• Retrieve node hardware specs for resource-aware task scheduling
• Compress and decompress data on the fly as part of a pipeline

FAQ from mcp-server

What capabilities does mcp-server offer?

It provides HDF5 file reading/listing, Slurm job submission/status checks, node hardware queries (CPU, memory, disk), and string/file compression with gzip or zlib.

How do I install and run mcp-server?

Use uv to create a virtual environment, install the package with uv pip install -e ., then run python -m src.server. The server starts on http://localhost:8000.

What endpoints does mcp-server expose?

It exposes POST /mcp for MCP requests and GET /health for health checks.

What dependencies or runtime are required?

Python with uv for dependency management. The server uses packages listed in pyproject.toml (installable via uv pip install -e .). No specific Python version is mentioned.

Does mcp-server support authentication or transports other than HTTP?

No authentication is mentioned. The only transport is HTTP at http://localhost:8000.