MCPScan

This repository is archived. I started this as an experiment to work with Aider, but did not make enough progress to make it useful. It will currently clone a repo in a docker container, then run semgrep rules and dependancy scans. I did not do enough testing or output formatting to rely on this, but it may be a good starting point for someone else.

A specialized security scanning tool for Model Context Protocol (MCP) servers. MCPScan performs comprehensive security analysis of MCP server implementations using multiple scanning tools:

• Semgrep for code pattern analysis
• npm audit for JavaScript/Node.js dependencies
• pip-audit for Python dependencies

Features

• Automated MCP server repository cloning and scanning
• Multi-tool security analysis tailored for MCP servers:
  • Static code analysis with Semgrep rules for:
    • Dangerous code patterns that could compromise model context
    • Local file access vulnerabilities
    • Network access security
    • Obfuscated code detection
    • Process execution monitoring
    • HTTP/HTTPS endpoint analysis
  • Dependency vulnerability scanning:
    • Python package vulnerabilities via pip-audit
    • JavaScript package vulnerabilities via npm audit
• Automatic MCP server framework detection
• Results aggregation and reporting in JSON format
• Docker containerization for isolated scanning
• Automatic cleanup of temporary files

Prerequisites

• Docker installed and running
• Python 3.x (for running MCP-Get scanner)
• Internet connection for repository cloning and package list fetching

Installation

1. Clone this repository
2. Build the Docker container:

./src/docker_build.sh

Usage

Scanning a Single Repository

./src/docker_run_one.sh <repository-url>

Example:

./src/docker_run_one.sh "https://github.com/modelcontextprotocol/servers"

Scanning All Servers in the MCP Get repo

python3 src/docker_run_mcp_get.py

This will:

1. Fetch the MCP server list from MCP-Get
2. Clone each MCP server repository
3. Run comprehensive security scans
4. Save detailed analysis to the results directory

Output

Results are processed through multiple stages:

1. Individual scan results are saved to the results directory:
   • Semgrep analysis results
   • Package vulnerability scans (pip-audit/npm audit)
2. Results are combined into a single JSON file in results/combined
3. Final reduced results in results/reduced:
   • JSON summary with findings by rule
   • Detailed vulnerability information
   • Human-readable text report
   • Simplified format for easy parsing

The reduced results include:

• Total findings count
• Findings categorized by rule type
• Dependencies scan summary with vulnerability counts
• Detailed vulnerability information for each package
• Code analysis findings with file locations and snippets

Project Structure

• src/docker/semgrep_rules/ - Custom Semgrep rule definitions
• src/docker/ - Core scanning logic and utilities
  • package_scan.py - Dependency vulnerability scanning
  • cleanup.py - Temporary file management
  • Other scanning utilities
• results/ - Scan output directory (created during execution)

Dependencies

This project relies on:

• Docker
• Python 3.x
• Semgrep (installed in Docker container)
• pip-audit (installed during scanning)
• npm (for JavaScript projects)
• Requests library for Python

Third-Party Attributions

• Semgrep - Static analysis tool (OSS License)
• pip-audit - Python dependency scanner (Apache 2.0)
• npm audit - Node.js dependency scanner
• Requests - HTTP library for Python (Apache 2.0)
• MCP-Get - Package list source

License

This project is licensed under the Mozilla Public License Version 2.0. See the LICENSE file for details.

Contributing

[Add contribution guidelines here]

Output Structure

Scan results are organized in three stages:

1. Individual scan results in results/
2. Combined results in results/combined/
3. Reduced results in results/reduced/ containing:
   • Summary of findings by rule type
   • Detailed vulnerability information
   • Simplified findings format
   • Human-readable text report

TODO

• Reduce the output jsons to a single representation
• Add support for go
• Add result caching, store last tested hash for a repo
• More tests and scans
• Add severity scoring system
• Implement parallel scanning for multiple repositories