convez.ai
MCP ServerDevOps & InfraestruturaIntermediário
Naidu2404

Argus CI — Agente de Segurança em Código

Naidu2404/argus-ci

Segurança de código automática para Cursor e Claude com Semgrep

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP server que integra análise de segurança com Semgrep em seu fluxo de desenvolvimento. Escaneia código gerado por IA em tempo real no editor, bloqueia commits inseguros via pre-commit hook, e oferece reviews conversacionais de PRs. Detecta vulnerabilidades OWASP Top 10, injeção SQL, XSS, secrets hardcoded e mais.

Instalação

$npx convez-ai Naidu2404/argus-ci

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
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Bloquear commits com vulnerabilidades conhecidas antes de fazer push
  • Revisar PRs sob aspecto de segurança via chat conversacional
  • Validar código gerado por IA em tempo real no editor antes de usar
  • Detectar secrets, injeção SQL, XSS e OWASP Top 10 automaticamente

Veredito

Ferramenta sólida que integra análise de segurança estática no fluxo de desenvolvimento IA-first, com setup zero-config e detecção pragmática de vulnerabilidades comuns. Recomendado para equipes que usam Cursor/Claude e querem gates de segurança confiáveis antes de deploy.

README

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

argus-ci

AI-powered code security agent — catches vulnerabilities as your AI agent writes code, blocks bad commits, and reviews PRs on demand.

Powered by Semgrep under the hood. Works with Cursor, Claude Code, or any MCP-compatible editor.


What it does

  • In-IDE scan — auto-triggers after every AI code generation via MCP, surfaces issues before you even see the code
  • Pre-commit gate — mandatory Semgrep scan on staged files before every commit; errors block the commit
  • PR / branch review — conversational agent: type "review PR #142" and get a full security report
  • Zero config — auto-detects your stack (React, Vue, Node, Python, Go…) and picks the right rulesets

Catches: injection, XSS, hardcoded secrets, insecure crypto, path traversal, prototype pollution, OWASP Top 10, and more.


Requirements

  • Node.js ≥ 18
  • ANTHROPIC_API_KEY — only needed for the conversational agent (argus-ci chat)

Semgrep is installed automatically by npx argus-ci setup. No manual install needed.


1. Add to your AI editor (MCP)

Open Cursor Settings → MCP and add:

{
  "argus-ci": {
    "command": "npx",
    "args": ["argus-ci"]
  }
}

Claude Code — add to ~/.claude/settings.json:

{
  "mcpServers": {
    "argus-ci": {
      "command": "npx",
      "args": ["argus-ci"]
    }
  }
}

The MCP server registers as "argus" in Cursor's tool panel.

MCP tools available

ToolDescription
scan_filesScan specific files — called automatically after code generation
scan_stagedScan all git-staged files
scan_branchScan changed files on a branch vs base
scan_prScan a GitHub PR by URL, optionally post results as a comment

2. Run setup in your repo

One command does everything — installs Semgrep, copies AI trigger files, and installs the pre-commit hook.

cd your-repo
npx argus-ci setup

Output:

🚀 argus-ci setup

  ⚙️  Semgrep not found — installing automatically...
     → brew install semgrep
  ✓ Semgrep installed (semgrep 1.x.x)
  ✓ CLAUDE.md written
  ✓ .cursorrules written
  ✓ Pre-commit hook installed

✅ Setup complete. argus-ci is now active in this repo.

  What happens next:
  • Every file your AI agent writes is scanned automatically (via MCP)
  • Every commit is scanned — errors block the commit
  • CLAUDE.md and .cursorrules tell your AI agent to run scans automatically

  To review a PR:     npx argus-ci pr <github-url>
  To remove the hook: npx argus-ci setup --remove

The setup does three things automatically:

  • Semgrep — installed via Homebrew on macOS, pip3 elsewhere. Skipped if already installed.
  • CLAUDE.md / .cursorrules — copied into the repo root. Tell the AI agent to call scan_files after every code generation.
  • Pre-commit hook — written to .git/hooks/pre-commit. Blocks commits with ERROR-severity findings.

To remove: npx argus-ci setup --remove
Emergency bypass: git commit --no-verify (not recommended)


3. Conversational agent

Review a PR or branch in plain English. Requires ANTHROPIC_API_KEY.

export ANTHROPIC_API_KEY=sk-ant-...

# One-shot PR review
argus-ci pr https://github.com/org/repo/pull/142

# Interactive REPL
argus-ci chat
# You: review PR https://github.com/org/repo/pull/142
# You: check branch feature/payments
# You: what issues are in my current changes

4. GitHub Actions (CI gate)

Add to .github/workflows/argus-ci.yml:

name: argus-ci security scan

on:
  pull_request:
    branches: [main, develop]

jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0

      - uses: actions/setup-node@v4
        with:
          node-version: 20

      - name: Install Semgrep
        run: pip install semgrep

      - name: Run argus-ci
        run: npx argus-ci scan --branch ${{ github.head_ref }}
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

Rulesets

Auto-detected from your project. No config needed.

RulesetWhen used
p/secretsAlways — catches hardcoded API keys, tokens, passwords
p/owasp-top-tenAlways — injection, XSS, broken auth, etc.
p/javascriptAny JS project
p/typescriptTypeScript detected
p/reactReact dependency found
p/nodejsExpress/Fastify/NestJS/etc. found
p/nextjsNext.js detected
p/pythonPython project
p/golangGo project
p/javaJava/Kotlin project

Environment variables

VariableRequiredPurpose
ANTHROPIC_API_KEYFor chat/agentPowers the conversational interface
GITHUB_TOKENFor private reposFetch PR files, post PR comments

License

MIT © Venkat Swara Moyya

Ver o original no GitHub

Bom para

Relacionados