praqtor-mcp 1.0.0

PRAQTOR MCP-S — AI Agent Security Scanner for MCP servers

PRAQTOR MCP-S

AI Agent Security Scanner for MCP Servers

Scan live MCP servers and configuration files for security vulnerabilities. 8 detection engines, attack chain analysis, zero false positives.

Installation

pip install praqtor-mcp

Quick Start

# Set your API key
export PRAQTOR_API_KEY=your-api-key

# Scan a live MCP server by URL
praqtor-mcp scan-url https://mcp.example.com/sse

# Scan a config file
praqtor-mcp scan config.json

# Output as JSON
praqtor-mcp scan-url https://mcp.example.com/sse --format json

# Output as SARIF (for GitHub Advanced Security)
praqtor-mcp scan-url https://mcp.example.com/sse --format sarif > results.sarif

How It Works

URL Scan — Point at any live MCP server. PRAQTOR connects via SSE or Streamable HTTP, discovers tools automatically, then runs all 8 security engines against the real tool definitions.

Config Scan — Pass your mcpServers JSON configuration. PRAQTOR analyzes the server definitions, tool schemas, and cross-server interactions.

Detection Engines

Engine	Code	What It Detects
Secret & Data Exposure	SDE	API keys, tokens, credentials in configs
Semantic Threat Analysis	STA	Tool poisoning, description manipulation
Schema Integrity Monitor	SIM	Schema drift, type confusion attacks
Package Provenance	PPV	Supply chain risks, unverified packages
Cross-Origin Influence	COI	Cross-server trust violations
Data Flow Path Analysis	DFP	Sensitive data flowing between tools
Permission Alignment	CPA	Over-permissioned tool capabilities
Attack Surface Hardening	ASH	Injection surfaces, SSRF, command injection

Plus ENABLES — attack chain wiring that connects findings across engines.

Commands

scan-url — Scan a Live MCP Server

praqtor-mcp scan-url https://mcp.example.com/sse

Connects to the MCP server, performs protocol handshake, discovers all tools via tools/list, and runs the full 8-engine scan.

scan — Scan a Config File

praqtor-mcp scan my-mcp-config.json

Analyzes an MCP configuration file (the same format used by Claude Desktop, Cursor, etc.).

health — Check API Status

praqtor-mcp health

Output Formats

Format	Flag	Use Case
Text	--format text	Terminal output with colors (default)
JSON	--format json	Programmatic consumption, CI/CD
SARIF	--format sarif	GitHub Advanced Security integration

Filtering

# Only show critical findings
praqtor-mcp scan-url https://mcp.example.com/sse --severity critical

# Limit output
praqtor-mcp scan-url https://mcp.example.com/sse --limit 10

CI/CD Integration

GitHub Actions

- name: Scan MCP Config
  run: |
    pip install praqtor-mcp
    praqtor-mcp scan mcp-config.json --format sarif > praqtor-results.sarif
  env:
    PRAQTOR_API_KEY: ${{ secrets.PRAQTOR_API_KEY }}

- name: Upload SARIF
  uses: github/codeql-action/upload-sarif@v3
  with:
    sarif_file: praqtor-results.sarif

Exit Codes

Code	Meaning
0	Passed — no critical or high findings
1	Warning — high severity findings detected
2	Failed — critical findings detected

Configuration

Set credentials via environment variables or command-line flags:

# Environment variables (recommended)
export PRAQTOR_API_URL=https://praqtormcp.fly.dev
export PRAQTOR_API_KEY=your-api-key

# Or use flags
praqtor-mcp scan-url https://example.com/sse --api-url https://praqtormcp.fly.dev --api-key YOUR_KEY

API

PRAQTOR MCP-S also provides a REST API for programmatic access:

curl -X POST https://praqtormcp.fly.dev/api/v1/scan/url \
  -H "X-API-Key: YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://mcp.example.com/sse"}'

Full API docs: https://praqtormcp.fly.dev/api

License

MIT