convez.ai
MCP Server
test008008008008-glitch

mcp-cateye

test008008008008-glitch/mcp-cateye

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP Server security scanner with fuzzing — the first tool that actually attacks MCP servers to find vulnerabilities, not just reads their config files.

Instalação

$npx convez-ai test008008008008-glitch/mcp-cateye

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

🐱 mcp-cateye

MCP Server Security Scanner — fuzzing + static analysis. Cat's eye sees what others miss.

English | 中文

PyPI Python License: MIT

Features

Dynamic Analysis (Fuzzing)

  • Active fuzzing of MCP server tools with 50+ payloads across 9 categories
  • Command injection, path traversal, SSRF, prompt injection, SQL injection, XSS, SSTI, info disclosure, deserialization
  • Category filtering for targeted testing
  • JSON output for CI/CD integration

Static Analysis (NEW in v1.1.0)

  • Config Discovery — scan Claude, Cursor, VS Code, Windsurf, Zed, Cline, RooCode configs
  • Secrets Detection — find hardcoded API keys, tokens, passwords (OpenAI, GitHub, AWS, Slack, JWT, etc.)
  • Tool Description Analysis — detect poisoning, rug pull, prompt injection, over-privileged tools
  • Dependency Scanning — pip-audit integration for known CVEs
  • Readiness Checks — AST-based analysis for missing timeouts, shell=True, bare excepts
  • Security Score — 0-100 score with A-F grade breakdown
  • AI-BOM — CycloneDX 1.5 JSON bill of materials for MCP servers

Why mcp-cateye?

graph LR
    A[Active Fuzzing 🐱] --> E[mcp-cateye]
    B[Static Analysis] --> E
    C[Security Scoring] --> E
    D[AI-BOM Generation] --> E
    E --> F[One CLI. Zero blind spots.]

Existing MCP security tools each cover one or two of these. mcp-cateye is the only one that ships all four in a single CLI — so you can find the bugs that static-only scanners miss (command injection, path traversal, SSRF) and the bugs that fuzzers miss (hardcoded secrets, vulnerable dependencies, poisoned tool descriptions) in one scan.

Cat's eye sees what others miss. 🐱

Installation

From GitHub (recommended for now)

pip install git+https://github.com/test008008008008-glitch/mcp-cateye.git

From PyPI (coming soon)

pip install mcp-cateye

From source

git clone https://github.com/test008008008008-glitch/mcp-cateye
cd mcp-cateye
pip install -e .

Requires Python 3.10+

Quick Start

Fuzz an MCP server

# Fuzz all tools
mcp-cateye fuzz python -- -m my_mcp_server

# Fuzz specific categories
mcp-cateye fuzz -c cmd -c ssrf node -- server.js

# JSON output for CI
mcp-cateye fuzz --json python -- server.py

Static analysis

# Full scan
mcp-cateye scan .

# Quick security score
mcp-cateye score .

# Generate AI-BOM
mcp-cateye scan . --aibom

# Scan specific clients
mcp-cateye scan . --clients claude --clients cursor

List tools and payloads

# List tools on a server
mcp-cateye list-tools python -- -m my_mcp_server

# List available payloads
mcp-cateye list-payloads

Security Score

The score command gives a quick 0-100 rating:

🐱 mcp-cateye — Security Score

  🟡 B+  [██████████████████████░░░░░░░░]  72/100

  Breakdown:
    fuzzing              [████████████████████] 100/100
    secrets              [██████████░░░░░░░░░░]  50/100
    dependencies         [████████████████████] 100/100
    tool_descriptions    [██████████████░░░░░░]  70/100
    readiness            [████████████████░░░░]  80/100

Payload Categories

CategoryCountExamples
Command Injection8$(whoami), backticks, os.system
Path Traversal6../../etc/passwd, null bytes
SSRF6http://169.254.169.254, DNS rebinding
Prompt Injection7Ignore instructions, role hijack
SQL Injection6Union, blind, time-based
XSS5Script tags, event handlers
Template Injection5Jinja2, Twig, Freemarker
Info Disclosure4Stack traces, debug endpoints
Deserialization4Pickle, YAML unsafe

CI/CD Integration

# GitHub Actions
- name: Security Scan
  run: |
    pip install mcp-cateye
    mcp-cateye scan . --json > report.json
    mcp-cateye score .

Exit codes:

  • 0 — No critical findings
  • 1 — Critical findings detected

License

MIT

Ver o original no GitHub

Bom para

Relacionados