convez.ai
MCP ServerSegurançaIntermediário
sulthonzh

MCP Audit

sulthonzh/mcp-audit

Auditoria de segurança para MCP servers, Docker, K8s e Helm

Ver no GitHub
InicialPluginScore · de 0 a 100

Scanner de segurança para servidores MCP (Model Context Protocol) e infraestrutura cloud-native. Analisa configurações MCP, Dockerfiles, manifests Kubernetes e Helm charts para detectar permissões perigosas, segredos hardcoded e vulnerabilidades antes de entrar em produção.

Instalação

$npx convez-ai sulthonzh/mcp-audit

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
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 1 mês
Versão
—
Issues abertas
1
Contribuidores
1

Casos de uso

  • Validar segurança de MCP servers antes de integrar com Claude
  • Auditar Dockerfiles e imagens em CI/CD pipelines
  • Verificar manifestos Kubernetes quanto a privilégios excessivos
  • Scanear Helm charts para configurações inseguras e segredos expostos
  • Integrar checks de segurança em workflows de deployment

Veredito

Ferramenta essencial para quem usa MCP servers em produção. Aborda um gap real — auditar o que plugins podem acessar — e expande o escopo para toda a stack (containers, orquestração, charts). Implementação sólida com suporte a CI/CD.

README

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

MCP Audit 🔍

Security scanner for MCP (Model Context Protocol) servers — and your container/K8s/Helm infrastructure too. Think of it as npm audit but for your AI agent integrations and cloud-native deployments.

Why MCP Audit?

Everyone's installing MCP servers like crazy, but nobody's checking if they're about to give a plugin full access to their filesystem, data, and context. MCP Audit gives you visibility into what these servers can actually do — and flags the risky stuff before it bites you.

It also scans Dockerfiles, Kubernetes manifests, and Helm charts, because your AI tools don't live in a vacuum. They run in containers, on clusters, behind charts. Might as well audit the whole stack.

What It Scans

TargetCommandWhat It Catches
MCP config filesmcp-audit scanRisky permissions, overly broad file access
MCP server reposmcp-audit check <repo>Prompt injection, hardcoded secrets, trust scoring
Dockerfilesmcp-audit docker <path>Root user, exposed secrets, outdated base images
Kubernetes manifestsmcp-audit k8s <path>Privileged containers, hostNetwork, runaway resources
Helm chartsmcp-audit helm <path>Hardcoded secrets in values.yaml, unsafe defaults

Quick Start

npm install -g mcp-audit

# Scan your MCP config
mcp-audit scan

# Audit a remote MCP server
mcp-audit check github.com/user/mcp-server

# Scan a Dockerfile
mcp-audit docker ./Dockerfile

# Scan K8s manifests
mcp-audit k8s ./manifests

# Scan a Helm chart
mcp-audit helm ./my-chart

# CI mode (exits with code on findings)
mcp-audit check --ci

Usage

Scan MCP Configuration

Checks claude_desktop_config.json, .cursor/mcp.json, and other MCP config files for risky permissions.

mcp-audit scan
mcp-audit scan -o report.json  # save report

Check a Remote Server

Clones the repo, runs static analysis, and generates a trust score based on GitHub signals.

mcp-audit check https://github.com/username/mcp-server
mcp-audit check https://github.com/username/mcp-server --ci  # CI-friendly

Docker Security

mcp-audit docker ./Dockerfile
mcp-audit docker ./docker-dir  # scans all Dockerfiles in directory

Detects: root user, ADD vs COPY, hardcoded secrets, latest tags, missing .dockerignore.

Kubernetes Security

mcp-audit k8s ./manifests
mcp-audit k8s ./manifests --strict  # stricter checks

Detects: privileged containers, hostNetwork/hostPID, missing resource limits, alwaysPullPolicy not set, containers running as root.

Helm Chart Security

mcp-audit helm ./my-chart
mcp-audit helm ./my-chart --strict -o report.json

Automatically detects Helm charts (looks for Chart.yaml). Scans values.yaml for hardcoded secrets and privileged flags, strips Go template syntax from templates/ and runs K8s security checks, and validates Chart.yaml for deprecated API versions and missing metadata.

CI Integration

All scanners support --ci for pipeline-friendly output and proper exit codes. Use --strict to fail on warnings too.

# GitHub Actions example
- name: Security Audit
  run: |
    npx mcp-audit k8s ./k8s --ci --strict
    npx mcp-audit docker . --ci
    npx mcp-audit helm ./charts --ci

Configuration

Create mcp-audit.config.json to customize:

{
  "vulnerabilityDatabase": "https://raw.githubusercontent.com/your-org/mcp-vuln-db/main/database.json",
  "trustWeight": {
    "stars": 0.3,
    "tests": 0.3,
    "ci": 0.2,
    "age": 0.2
  }
}

Pre-commit Hook

# .pre-commit-config.yaml
repos:
  - repo: local
    hooks:
      - id: mcp-audit
        name: MCP Security Audit
        entry: mcp-audit check --ci
        language: system

Contributing

PRs welcome. Open an issue first if it's a significant change.

License

MIT

Ver o original no GitHub

Bom para

Relacionados