convez.ai
MCP ServerSegurançaAvançado
mowen628

SIFT MCP — Agente Autônomo de Resposta a Incidentes

mowen628/sift-mcp

IR autônomo: triage de rede + forense profunda com auditoria estruturada

Ver no GitHub
InicialPluginScore · de 0 a 100

Servidor MCP que integra Claude Code com o SANS SIFT Workstation para análise de incidentes autônoma em tempo real. Combina triage de rede ao vivo (DNS, descoberta de dispositivos, logs de container) com forense profunda (análise de memória, varredura IOC, timeline), registrando tudo em auditoria estruturada para InfluxDB e Grafana.

Instalação

$npx convez-ai mowen628/sift-mcp

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
1

Casos de uso

  • Investigação automatizada de ameaças em redes domésticas/corporativas com ~50 dispositivos
  • Análise forense de memória, IOC scanning e detecção de código injetado sem intervenção manual
  • Pipeline completo triage→forense→relatório orquestrado por Claude Code com trilha de auditoria

Veredito

Ferramenta especializada e bem arquitetada para IR autônomo, com guardrails em camada MCP (não apenas prompt), integração real com SIFT/Volatility3/YARA e auditoria estruturada. Excelente para equipes de segurança que querem automação confiável; requer expertise em forense e acesso a infraestrutura de triage.

README

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

sift-mcp

Autonomous incident response agent built on the SANS SIFT Workstation and Protocol SIFT. Submitted to the Find Evil! hackathon (Apr 15 – Jun 15, 2026).

What It Does

sift-mcp is a custom MCP server that gives Claude Code autonomous IR capabilities across two layers:

LayerToolsData Sources
Live Network TriageDNS log analysis, device discovery, container log triageAdGuard Home API, nmap, Docker SSH
Deep ForensicsMemory analysis, IOC scanning, timeline generationVolatility3, YARA, Plaso, EZ Tools

Findings flow through a structured audit trail into InfluxDB and Grafana, with optional Home Assistant alerts for high-severity events. The server runs as a persistent systemd service on the SIFT VM, analyzing a real home network with ~50 devices.

Architecture

Claude Code (orchestrator)
        │
        │ stdio (MCP protocol)
        ▼
┌─────────────────────────────────────────┐
│              server.py                  │
│         MCP Tool Registry               │
├──────────────┬──────────────────────────┤
│  constraints.py  │  audit.py            │
│  (guardrails)    │  (structured log)    │
├──────────────┴──────────────────────────┤
│              tools/                     │
│  network.py  logs.py  memory.py         │
│  ioc.py      timeline.py  case.py       │
└─────────────────────────────────────────┘
        │                    │
        ▼                    ▼
  Live Network          Deep Forensics
  (AdGuard, nmap,       (Volatility3,
   Docker logs)          YARA, Plaso)
        │
        ▼
  InfluxDB (sift-ir bucket) → Grafana dashboard
  Home Assistant webhook (high-severity alerts)

See docs/architecture.md for full detail.

Judging Criteria Alignment

CriterionImplementation
Autonomous Execution QualityClaude Code orchestrates full IR pipeline without human intervention; tools chain automatically from triage → forensics → report
IR AccuracyReal network data — live AdGuard DNS logs, actual device inventory, real memory/disk artifacts
Breadth & Depth6 tool modules spanning live triage, memory forensics, IOC scanning, timeline analysis, and case management
Constraint ImplementationArchitectural guardrails in constraints.py: read-only by default, path allowlist, no arbitrary exec, explicit confirmation for destructive ops — enforced at the MCP layer, not prompt-based
Audit Trail QualityStructured JSON logs via audit.py for every tool call: timestamp, tool, args, result summary, severity
Usability / DocumentationTry-it-out instructions below; CLAUDE.md for AI-assisted operation; Grafana dashboard for findings visibility

Tools

ToolModuleDescription
network_dns_querynetwork.pyQuery AdGuard DNS logs for a host/IP
network_device_scannetwork.pynmap sweep, identify unknown devices
logs_containerlogs.pyTail/search Docker container logs via SSH
memory_pslistmemory.pyList processes from memory dump (Volatility3)
memory_netscanmemory.pyNetwork connections from memory
memory_malfindmemory.pyFind injected code in memory
ioc_yara_scanioc.pyYARA scan file or directory
ioc_hashioc.pyHash a file (md5/sha1/sha256)
timeline_createtimeline.pyRun log2timeline on an artifact
case_createcase.pyInitialize a new case
case_add_findingcase.pyRecord a finding with severity
case_reportcase.pyGenerate case report, write to InfluxDB

Requirements

  • SANS SIFT Workstation (Ubuntu 24.04)
  • Python 3.12+ with venv
  • Claude Code (claude-code CLI)
  • SSH access to Docker host (for container log triage)

Try It Out

See docs/try-it-out.md for full setup and usage instructions.

Quick start:

git clone https://github.com/mowen628/sift-mcp.git ~/sift-mcp
cd ~/sift-mcp
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # fill in your values
# Register MCP server with Claude Code
claude mcp add sift-mcp -- python /path/to/sift-mcp/server.py
claude  # start Claude Code — sift-mcp tools are now available

Deliverables

  • GitHub repository
  • 5-minute demo video
  • Architecture diagram (docs/architecture.md)
  • Devpost writeup
  • Dataset documentation (docs/dataset.md)
  • Accuracy report (docs/accuracy-report.md)
  • Try-it-out instructions (docs/try-it-out.md)
  • Agent execution logs (logs/)

License

MIT — see LICENSE

Ver o original no GitHub

Bom para

Relacionados