AI Detection Engineering Platform

License: MIT Python 3.11+ MCP 2025-11-25 Docker

An AI-driven detection-engineering platform for Wazuh. Investigate alerts, hunt threats, tune false positives, edit and validate detection rules, and run Atomic Red Team validations across your Wazuh deployment — by talking to any AI assistant. Built as a Model Context Protocol (MCP) server.

v4.2.1 | 48 security tools | Wazuh 4.8.0–4.14.4 | Changelog

---

What This Does

Your Wazuh SIEM generates thousands of alerts, vulnerability findings, and agent events daily. Investigating them means juggling dashboards, writing API queries, and manually correlating data across tools.

This MCP server turns that workflow into a conversation:

You:    "Show me critical alerts from the last hour"
AI:     [calls get_wazuh_alerts] Found 3 critical alerts:
        1. SSH brute force from 10.0.1.45 → agent-003 (Rule 5712, Level 10)
        2. Rootkit detection on agent-007 (Rule 510, Level 12)
        3. FIM change /etc/shadow on agent-001 (Rule 550, Level 10)

You:    "Block that source IP on agent-003"
AI:     [calls wazuh_block_ip] Blocked 10.0.1.45 via firewall-drop on agent-003.

You:    "Which agents have unpatched critical CVEs?"
AI:     [calls get_critical_vulnerabilities] 3 agents with critical vulnerabilities...

It works with Claude Desktop, Open WebUI + Ollama (fully local, air-gapped), mcphost, or any MCP-compliant client.

---

Works With Cloud AND Local LLMs

This is a standard MCP tool server. It doesn't care what LLM you use — it just executes tools and returns results.

Mode	LLM	Client	Data leaves your network?
Cloud	Claude, GPT, etc.	Claude Desktop, any MCP client	Yes (to LLM provider)
Local	Llama, Qwen, Mistral via Ollama	Open WebUI, mcphost, IBM/mcp-cli	No. Fully air-gappable.

For security teams that can't send SIEM data to cloud APIs (compliance, air-gapped networks, data sovereignty), the local mode with Ollama keeps everything on-premises. Both modes coexist — same server, same tools, same API.

Quick Start: Local LLM with mcphost

# 1. Start the MCP server
docker compose up -d

# 2. Install mcphost (Go binary, no dependencies)
go install github.com/mark3labs/mcphost@latest

# 3. Configure
cat > ~/.mcphost.yml << 'EOF'
mcpServers:
  wazuh:
    type: remote
    url: http://localhost:3000/mcp
    headers: ["Authorization: Bearer ${env://MCP_API_KEY}"]
EOF

# 4. Chat with your SIEM using a local model
export MCP_API_KEY="your-key-from-server-logs"
mcphost --model ollama/qwen2.5:7b

Quick Start: Multi-User SOC with Open WebUI

Open WebUI v0.6.31+ connects to our /mcp endpoint natively. Add it as an MCP tool server in Admin Settings, and your entire team gets AI-powered SIEM analysis with conversation history, RBAC, and a web UI.

---

48 Security Tools

Every tool is validated, rate-limited, scope-checked, and audit-logged.

Category	Tools	What They Do
Alerts (4)	get_wazuh_alerts get_wazuh_alert_summary analyze_alert_patterns search_security_events	Query, filter, search, and analyze alert data via Elasticsearch
Agents (6)	get_wazuh_agents get_wazuh_running_agents check_agent_health get_agent_processes get_agent_ports get_agent_configuration	Monitor agent status, running processes, open ports, and configs
Vulnerabilities (3)	get_wazuh_vulnerabilities get_critical_vulnerabilities vulnerability_summary	Query CVEs by severity, agent, and package
Security Analysis (6)	analyze_security_threat check_ioc_reputation perform_risk_assessment get_top_security_threats generate_security_report run_compliance_check	Threat analysis, IOC lookup, risk scoring, compliance checks
System (10)	get_wazuh_statistics get_wazuh_cluster_health get_wazuh_rules_summary search_wazuh_manager_logs ...	Cluster health, rules, manager logs, stats
Active Response (9)	wazuh_block_ip wazuh_isolate_host wazuh_kill_process wazuh_disable_user wazuh_quarantine_file ...	Block IPs, isolate hosts, kill processes, quarantine files
Verification (5)	wazuh_check_blocked_ip wazuh_check_agent_isolation wazuh_check_process wazuh_check_user_status ...	Verify active response actions took effect
Rollback (5)	wazuh_unisolate_host wazuh_enable_user wazuh_restore_file wazuh_firewall_allow wazuh_host_allow	Undo active response actions

---

Quick Start

Prerequisites

• Docker 20.10+ with Compose v2
• Wazuh 4.8.0–4.14.4 with API access enabled

Deploy

git clone https://github.com/gensecaihq/Wazuh-MCP-Server.git
cd Wazuh-MCP-Server
cp .env.example .env

Edit .env:

WAZUH_HOST=your-wazuh-server
WAZUH_USER=your-api-user
WAZUH_PASS=your-api-password

docker compose up -d
curl http://localhost:3000/health

Connect Claude Desktop

1. Settings → Connectors → Add custom connector
2. URL: https://your-server/mcp
3. Add Bearer token in Advanced settings

Detailed setup: Claude Integration Guide

---

Security

This server sits between an LLM and your SIEM. Security is not optional.

Layer	What It Does
RBAC	Per-tool scope enforcement. 14 active response tools require wazuh:write. Read-only tokens can query but never trigger actions. Authless mode is read-only by default.
Audit Logging	Every destructive tool call (block IP, isolate host, kill process) is logged with client ID, session, timestamp, and full arguments.
Output Sanitization	Credentials, tokens, and API keys in alert full_log fields are redacted before reaching the LLM. Prevents credential leakage through AI responses.
Input Validation	Every parameter validated: regex agent IDs, ipaddress module for IPs, shell metacharacter blocking for active response, Elasticsearch Query DSL (no string interpolation).
Rate Limiting	Per-client sliding window with escalating block duration (10s → 5min).
Circuit Breakers	Wazuh API failures trigger fail-fast for 60s, auto-recover. Single trial in HALF_OPEN state.
Log Sanitization	Global filter redacts passwords, tokens, secrets from all server logs.
Container Hardening	Non-root user, read-only filesystem, CAP_DROP ALL, no-new-privileges.

# Generate a secure API key
python -c "import secrets; print('wazuh_' + secrets.token_urlsafe(32))"

---

Configuration

Required

Variable	Description
WAZUH_HOST	Wazuh Manager hostname or IP
WAZUH_USER	API username
WAZUH_PASS	API password

Optional

Variable	Default	Description
WAZUH_PORT	55000	Manager API port
MCP_HOST	0.0.0.0	Server bind address
MCP_PORT	3000	Server port
AUTH_MODE	bearer	oauth, bearer, or none
AUTH_SECRET_KEY	auto-generated	JWT signing key
AUTHLESS_ALLOW_WRITE	false	Allow active response in authless mode
ALLOWED_ORIGINS	https://claude.ai	CORS origins (comma-separated)
REDIS_URL	—	Redis URL for multi-instance session storage

Wazuh Indexer (for alert search + vulnerabilities)

Variable	Default	Description
WAZUH_INDEXER_HOST	—	Indexer hostname
WAZUH_INDEXER_PORT	9200	Indexer port
WAZUH_INDEXER_USER	—	Indexer username
WAZUH_INDEXER_PASS	—	Indexer password

Full reference: Configuration Guide

---

API Endpoints

Endpoint	Method	Description
/mcp	POST/GET/DELETE	MCP Streamable HTTP (recommended)
/sse	GET	Legacy Server-Sent Events
/health	GET	Health check (no auth required)
/metrics	GET	Prometheus metrics
/auth/token	POST	Exchange API key for JWT
/docs	GET	OpenAPI documentation

---

Architecture

src/wazuh_mcp_server/
├── server.py           # MCP protocol + 48 tool handlers
├── config.py           # Environment-based configuration
├── auth.py             # JWT + API key authentication
├── oauth.py            # OAuth 2.0 with Dynamic Client Registration
├── security.py         # Rate limiting, CORS, input validation
├── monitoring.py       # Prometheus metrics, structured logging
├── resilience.py       # Circuit breakers, retries, graceful shutdown
├── session_store.py    # Pluggable sessions (in-memory + Redis)
└── api/
    ├── wazuh_client.py    # Wazuh Manager REST API client
    └── wazuh_indexer.py   # Wazuh Indexer (Elasticsearch) client

---

Take It Further: Autonomous Agentic SOC

Combine this MCP server with Wazuh OpenClaw Autopilot to build a fully autonomous Security Operations Center.

While this server gives you conversational access to Wazuh, OpenClaw deploys AI agents that work around the clock — triaging alerts, correlating incidents, and recommending responses without human intervention.

Manual SOC:    Alert → Analyst reviews → Hours → Response
Agentic SOC:   Alert → AI triages → Seconds → Response ready for approval

Explore OpenClaw Autopilot

---

Documentation

Guide	Description
Claude Integration	Claude Desktop setup and authentication
Configuration	Full configuration reference
Advanced Features	HA, serverless, compact mode
API Documentation	Per-tool documentation
Security	Security hardening guide
Troubleshooting	Common issues and solutions
Operations	Deployment, monitoring, maintenance

---

Contributing

We welcome contributions. See Issues for bugs and feature requests, Discussions for questions.

---

License

MIT

---

Acknowledgments

• Wazuh — Open source security platform
• Model Context Protocol — AI tool integration standard
• Ollama — Local LLM inference
• Open WebUI — Self-hosted AI chat interface
• mcphost — MCP CLI host with LLM support

---

Contributors

Contributors

Avatar	Username	Contributions
	@actions-user	💻 Code
	@Simoon896	💻 Code

Legend: 💻 Code · 🐛 Issues · 🔀 Pull Requests · 💬 Discussions

Auto-updated by GitHub Actions