SIFT MCP — Agente Autônomo de Resposta a Incidentes
mowen628/sift-mcp
IR autônomo: triage de rede + forense profunda com auditoria estruturada
Ver no GitHubServidor 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-mcp1 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
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:
| Layer | Tools | Data Sources |
|---|---|---|
| Live Network Triage | DNS log analysis, device discovery, container log triage | AdGuard Home API, nmap, Docker SSH |
| Deep Forensics | Memory analysis, IOC scanning, timeline generation | Volatility3, 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
| Criterion | Implementation |
|---|---|
| Autonomous Execution Quality | Claude Code orchestrates full IR pipeline without human intervention; tools chain automatically from triage → forensics → report |
| IR Accuracy | Real network data — live AdGuard DNS logs, actual device inventory, real memory/disk artifacts |
| Breadth & Depth | 6 tool modules spanning live triage, memory forensics, IOC scanning, timeline analysis, and case management |
| Constraint Implementation | Architectural 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 Quality | Structured JSON logs via audit.py for every tool call: timestamp, tool, args, result summary, severity |
| Usability / Documentation | Try-it-out instructions below; CLAUDE.md for AI-assisted operation; Grafana dashboard for findings visibility |
Tools
| Tool | Module | Description |
|---|---|---|
network_dns_query | network.py | Query AdGuard DNS logs for a host/IP |
network_device_scan | network.py | nmap sweep, identify unknown devices |
logs_container | logs.py | Tail/search Docker container logs via SSH |
memory_pslist | memory.py | List processes from memory dump (Volatility3) |
memory_netscan | memory.py | Network connections from memory |
memory_malfind | memory.py | Find injected code in memory |
ioc_yara_scan | ioc.py | YARA scan file or directory |
ioc_hash | ioc.py | Hash a file (md5/sha1/sha256) |
timeline_create | timeline.py | Run log2timeline on an artifact |
case_create | case.py | Initialize a new case |
case_add_finding | case.py | Record a finding with severity |
case_report | case.py | Generate 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