MCPAudit

Offensive security testing framework for Model Context Protocol (MCP) servers.

Make MCP security testing as easy as running a linter.

mcpaudit scan ./server.py

Problem

MCP servers expose databases, APIs, file systems, cloud resources, and SaaS tools to AI agents — often without rigorous security review. MCPAudit helps teams find issues before attackers do.

Features

| Module | Status | Description | |--------|--------|-------------| | Permission Analyzer | ✅ Alpha | Flags destructive and over-privileged tools | | Prompt Injection Simulator | ✅ Alpha | Tests known injection attack patterns | | Tool Abuse Testing | ✅ Alpha | Detects path traversal and misuse surfaces | | Data Leakage Detection | ✅ Alpha | Scans for secrets and sensitive references | | Agent Jailbreak Testing | 🚧 Planned | Resistance scoring against jailbreak suites | | Multi-Step Attack Chains | ✅ Alpha | Identifies dangerous tool combinations | | Risk Scoring Engine | ✅ Alpha | CVSS-inspired security score (0–100) | | Compliance Checks | ✅ Alpha | OWASP LLM Top 10 & MCP best practices | | CI/CD Integration | 🚧 Planned | GitHub Action for pipeline gates | | HTML Reports | ✅ Alpha | mcpaudit report → security-report.html | | MCP Fuzzer | 🔮 Roadmap | mcpaudit fuzz | | MCPAudit Agent | 🔮 Roadmap | mcpaudit pentest |

Quick Start

Prerequisites

Python 3.11+
uv (recommended)

Install

git clone https://github.com/hello-args/MCPVault.git
cd MCPVault
uv sync --all-extras

Scan an MCP server

uv run mcpaudit scan examples/vulnerable-mcp-server/server.py

Save JSON results and generate HTML:

uv run mcpaudit scan examples/vulnerable-mcp-server/server.py -o report.json
uv run mcpaudit report report.json -o security-report.html

CI gate (fail on critical)

uv run mcpaudit scan ./server.py --fail-on-critical

Architecture

           ┌──────────────┐
           │ MCP Server   │
           └──────┬───────┘
                  │
                  ▼
         ┌─────────────────┐
         │ MCPAudit Scanner  │
         └─────────────────┘
                  │
     ┌────────────┼────────────┐
     ▼            ▼            ▼
Permission   Injection     Leakage
Analyzer      Engine       Scanner
     ▼            ▼            ▼
       Risk Scoring Engine
                  ▼
          Security Report

Project Structure

MCPAudit/
├── src/mcpaudit/          # Main package (src layout)
│   ├── cli/             # Typer CLI (`scan`, `report`, `fuzz`, `pentest`)
│   ├── core/            # Scanner orchestration
│   ├── analyzers/       # Security analyzers
│   ├── scoring/         # Risk scoring engine
│   ├── compliance/      # OWASP & MCP compliance checks
│   ├── reporting/       # Models & HTML reports
│   └── mcp/             # MCP client & discovery
├── tests/               # pytest suite
├── examples/            # Sample vulnerable MCP servers
├── action/              # GitHub Action (planned)
└── docs/                # Documentation

Development

uv sync --all-extras
uv run pytest
uv run ruff check src tests
uv run ruff format src tests
pre-commit install

Positioning

| Tool | Domain | |------|--------| | SonarQube | Code quality | | OWASP ZAP | Web security | | Trivy | Container security | | Semgrep | Static analysis | | MCPAudit | MCP security |

Roadmap

See docs/roadmap.md for the phased plan — category risk scoring, GitHub Action, SARIF, attack simulation, and more.

Contributing

See CONTRIBUTING.md.

License

Apache License 2.0 — see LICENSE.

Security

To report vulnerabilities, see SECURITY.md.

MCP Servers