Systems & network administration student in Belgium, focused on defensive security, homelab infrastructure, and AI tooling.
MCP Shield — a security audit framework for Model Context Protocol (MCP) servers, scanning them before installation for supply-chain attacks, prompt injection, tool poisoning, and rug pulls.
- 17 detectors across 3 surfaces (source code, MCP metadata, runtime delta)
- 359 tests, zero dependencies (Python stdlib only)
- Battle-tested on 31+ real-world MCP servers
pip install mcp-shield-audit· PyPI
Vulnerabilities I found with MCP Shield and reported to real-world MCP projects:
| Project | Finding |
|---|---|
| QuantGeekDev/docker-mcp | Command injection via shell=True on unsanitized input |
| 0xKoda/WireMCP | Command injection via unsanitized exec() parameters |
| gilby125/mcp-proxmox | TLS certificate verification disabled by default |
Defensive security · MCP / LLM security · Homelab (Proxmox, WireGuard) · Python · Automation