mds-mcp 0.1.3

Model Cache Server - MCP server for downloading and caching ML models and datasets

MDS-MCP: Model Cache Server

A Model Control Protocol (MCP) server for efficiently downloading and caching machine learning models and datasets from various sources like Hugging Face.

Features

• 🚀 Fast Model & Dataset Downloads: Efficient downloading and caching of ML models and datasets
• 🔄 Smart Caching: Automatically manages local cache to avoid re-downloading
• 🔐 Authentication Support: Handles gated models and private repositories
• 🌐 Multiple Providers: Currently supports Hugging Face with extensible architecture
• 📊 Monitoring: Built-in Sentry integration for error tracking
• 🛠 MCP Compatible: Works seamlessly with MCP clients

Installation

From PyPI (Recommended)

pip install mds-mcp

From Source

git clone https://github.com/yourusername/mds-mcp.git
cd mds-mcp
pip install -e .

Quick Start

1. Environment Setup

Create a .env file in your project directory:

# Optional: Hugging Face token for gated/private models
HF_TOKEN=your_huggingface_token_here

# Optional: Sentry configuration
SENTRY_DSN=your_sentry_dsn_here
SENTRY_ENVIRONMENT=production
SENTRY_SAMPLE_RATE=0.1

# Server configuration
MCP_HOST=0.0.0.0
MCP_PORT=9004
LOG_LEVEL=INFO

2. Start the Server

mds-mcp

Or run with Python:

python -m mds_mcp.main

3. Using the MCP Tools

The server provides two main tools:

Download Model

{
    "tool": "download_model",
    "arguments": {
        "provider": "huggingface",
        "identifier": "microsoft/DialoGPT-medium",
        "token": "optional_hf_token"
    }
}

Download Dataset

{
    "tool": "download_dataset",
    "arguments": {
        "provider": "huggingface",
        "identifier": "imdb",
        "name": "optional_config_name",
        "token": "optional_hf_token"
    }
}

Supported Providers

Hugging Face

• Models: All public and gated models from Hugging Face Hub
• Datasets: Complete dataset downloads with all splits
• Authentication: Supports access tokens for gated/private content

API Reference

Tools

download_model(provider, identifier, token=None)

Downloads a pretrained model to local cache.

Parameters:

• provider (str): Model source (e.g., 'huggingface')
• identifier (str): Model repository ID
• token (str, optional): Access token for private/gated models

Returns:

• Success: {'status': 'downloaded', 'path': '<cache_dir>'}
• Cached: {'status': 'cached', 'path': '<cache_dir>'}
• Auth Required: {'status': 'auth_required', 'access_url': '...', 'instructions': [...]}
• Error: {'status': 'error', 'message': '...'}

download_dataset(provider, identifier, name=None, token=None)

Downloads a dataset with all splits to local cache.

Parameters:

• provider (str): Dataset source (e.g., 'huggingface')
• identifier (str): Dataset repository name
• name (str, optional): Dataset config or task name
• token (str, optional): Access token for private/gated datasets

Returns: Same format as download_model.

Development

Setup Development Environment

git clone https://github.com/yourusername/mds-mcp.git
cd mds-mcp
pip install -e ".[dev]"

Running Tests

pytest

Code Formatting

black mds_mcp/
isort mds_mcp/

Type Checking

mypy mds_mcp/

Architecture

The server is built with a modular architecture:

• Core Server: FastMCP-based server handling MCP protocol
• Providers: Pluggable backends for different model sources
• Services: Business logic for caching, authentication, and downloads
• Logging: Structured logging with optional Sentry integration

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

• 📚 Documentation
• 🐛 Report Issues
• 💬 Discussions

Changelog

See CHANGELOG.md for details about version changes.