Model Safety MCP server

This is a small helper that lets your AI check machine learning model files for hidden dangers before you open them. It looks at files on your computer or files you can download from a link. It is handy for anyone who downloads models from the internet and wants a quick safety check first.

What this MCP server does

You tell your AI which model file or web link to check. The AI passes that to this helper. The helper runs safety scanners on the file, looking for risky formats and hidden code. Then it gives you back a list of findings, a severity level, and what to do next. You can also point it at a whole folder to check many files at once.

Click to zoom

What you can do with it

• Check a single model file on your computer
• Check a model file from a web link
• See which scanners are installed and ready
• Get a severity level and suggested next steps
• Look inside a file for risky packaging or hidden pickle code
• Scan a whole folder of model files and see which ones look risky

Try asking your AI

• “Use artifact_safety_report with path=C:/models/credit-risk.pth”
• “Use artifact_safety_report with url=https://huggingface.co/someone/somemodel/resolve/main/model.h5”
• “Use scan_directory with path=C:/models/downloads”
• “Use available_scanners to see what is installed”

What it gives back to you

You get back a report in the chat. It lists each finding with a severity, a category, and the evidence the scanner saw. It also gives you a highest severity level, a count of findings, and recommended actions. For a folder scan, you get a list of the risky files so you can focus on those first.

Before you start

Who it's for

Anyone who downloads machine learning models from the internet and wants a quick safety check before using them.