mcp-audit-cli 0.1.0

Automated security scanner and grading CLI for MCP servers.

MCPaudit

Automated security scanner + grading system for Model Context Protocol (MCP) servers.

MCPaudit gives teams a fast A-F risk grade, concrete findings, and machine-readable output before an MCP server is wired into an agent workflow. The free OSS CLI is the trust builder; the natural paid tier is registry monitoring, scheduled scans, historical grades, dashboards, and policy gates for CI/procurement review.

Quick Start

pip install mcp-audit-cli

mcp-audit scan ./path/to/mcp-server
mcp-audit scan https://example.com/mcp --format json
mcp-audit scan ./server --format sarif --output mcp-audit.sarif

The legacy mcpaudit command is kept as a compatibility alias for the original detector framework.

Why It Exists

MCP servers turn tools, files, prompts, and APIs into model-accessible capabilities. That makes security posture visible at the model layer: weak auth, broad file access, tool poisoning, leaked secrets, unsafe command execution, and permissive network settings can become agentic blast radius.

MCPaudit is built to make that risk obvious early.

Vulnerability Coverage

The repository includes a 30-class CVE-style threat model and an MVP scanner with 15 MCP-focused checks:

• unauthenticated HTTP transports
• wildcard CORS
• plaintext HTTP endpoints
• hardcoded secrets
• command injection sinks
• shell execution
• path traversal
• broad filesystem access
• unsafe deserialization
• prompt or tool poisoning markers
• over-broad tool descriptions
• missing rate limiting
• debug mode enabled
• server/version disclosure
• missing audit logging

Output Formats

• table / terminal: human-readable report with grade and findings
• json: stable JSON for dashboards and CI
• sarif: GitHub code scanning compatible output

Architecture

mcpaudit/
├── src/mcpaudit/           # Original Click/Rich detector framework
├── src/mcp_audit/          # Local-first CLI scanner and SARIF/JSON renderer
├── docs/threat-model.md    # Full vulnerability taxonomy narrative
├── docs/vulnerability-taxonomy.schema.json
├── docs/vulnerability-taxonomy.json
├── tests/                  # Detector and CLI scanner tests
└── pyproject.toml

Grading

Findings are weighted by severity and converted into an A-F grade:

• A: no material findings
• B: low risk
• C: moderate risk
• D: high risk
• F: critical or severe accumulated risk

Selling Angle

Suggested first buyer: security-conscious teams adopting MCP internally who need a quick risk signal before allowing new servers into agent environments.

Paid tier wedge:

• registry monitoring for public/private MCP servers
• scheduled scans and drift alerts
• historical grades per server
• team dashboards and SARIF ingestion
• policy gates for CI and procurement review

Development

python -m venv .venv
pip install -e ".[dev]"
pytest
ruff check .
mcp-audit scan tests --format sarif --output mcp-audit.sarif

Roadmap

• v0.1 - Bootstrap + threat model
• v0.2 - Core detection engine MVP
• v0.3 - CLI MVP with JSON and SARIF
• v0.4 - Package publishing and launch assets
• v0.5 - Cloud registry proof of concept

License

MIT