🔐 Wazuh SIEM × Claude Desktop — Conversational SOC via MCP

A complete walkthrough of my hands-on cybersecurity project integrating a SIEM with an AI assistant to enable natural-language security operations.

---

🎯 Project Overview

This project demonstrates how AI can augment SOC operations by connecting Wazuh SIEM to Claude Desktop through the MCP (Model Context Protocol), allowing a security analyst to query the SIEM in natural language and receive instant, structured analysis — without manually navigating dashboards.

Slack enrichment is handled both automatically (Wazuh native integration, level 7+ alerts) and on-demand (via Claude Desktop conversation).

What I accomplished:

• ✅ Deployed Wazuh Manager + Indexer on Ubuntu Server VM
• ✅ Connected Windows and Kali Linux agents to the Wazuh Manager
• ✅ Built a Python MCP server exposing 5 tools to Claude Desktop
• ✅ Enabled natural-language querying of alerts, agents, and vulnerabilities
• ✅ Configured automatic Slack notifications for critical alerts (level 7+)
• ✅ Added on-demand AI-enriched Slack reporting from Claude Desktop

---

🏗️ Architecture

[Physical Machine — Windows 10 — Claude Desktop]
        │
        │ stdio (local subprocess)
        ▼
  wazuh_mcp_server.py  ←── Python MCP Server
        │
        ├── HTTPS :55000 → Wazuh Manager API
        ├── HTTPS :9200  → Wazuh Indexer (OpenSearch)
        └── HTTPS        → Slack Webhook
        │
        ▼
[VM Ubuntu Server — Wazuh Manager + Indexer]
        ▲
        │ Logs / Heartbeat (port 1514)
        ├── [VM Windows — Wazuh Agent]
        └── [VM Kali    — Wazuh Agent]

---

🛠️ Technical Implementation

Environment Setup

Host OS       : Windows 10 (Claude Desktop + Python MCP Server)
VM 1          : Ubuntu Server — Wazuh Manager 4.11.2 + Wazuh Indexer
VM 2          : Windows 10   — Wazuh Agent
VM 3          : Kali Linux   — Wazuh Agent / Attack simulation
Hypervisor    : VMware

MCP Server — 5 Tools Built

Tool	Description
get_critical_alerts	Fetch alerts filtered by severity and time range
get_agents_status	List all agents with status, IP, OS, version
get_vulnerabilities	CVEs detected per agent, filtered by severity
generate_security_report	Aggregate alerts + agents + vulns into a full report
send_to_slack	Post AI-enriched analysis to the SOC Slack channel

Natural Language Queries (Examples)

"Show me critical alerts from the last 24 hours"
"Which agents have unpatched critical vulnerabilities?"
"Generate a security report with recommendations"
"Send the critical alerts to Slack with an analysis"

Slack Integration

Automatic — via Wazuh native integration in ossec.conf :

<integration>
  <name>custom-slack</name>
  <hook_url>https://hooks.slack.com/services/XXX/YYY/ZZZ</hook_url>
  <level>7</level>
  <alert_format>json</alert_format>
</integration>

On-demand — via send_to_slack MCP tool called by Claude Desktop, posting an AI-generated summary with alert details, severity color coding, and timestamp directly to the SOC channel.

Claude Desktop Configuration

{
  "mcpServers": {
    "wazuh": {
      "command": "python",
      "args": ["C:\\wazuh-mcp\\wazuh_mcp_server.py"]
    }
  }
}

Network Ports Required

sudo ufw allow 55000/tcp   # Wazuh Manager API
sudo ufw allow 9200/tcp    # Wazuh Indexer (OpenSearch)
sudo ufw allow 1514/tcp    # Agent communication

---

📊 Results & Findings

What Worked Well

• ✅ Full natural-language SIEM querying in seconds
• ✅ Real-time alert detection (SSH sudo usage, PAM sessions, brute-force attempts)
• ✅ AI-generated security reports with structured recommendations
• ✅ Dual Slack enrichment — automatic + on-demand from Claude Desktop
• ✅ Multi-agent visibility (Ubuntu Manager, Windows agent, Kali agent)

Key Lessons

• MCP bridges the gap between complex security tools and conversational AI
• Separating authentication systems (Wazuh API vs Indexer) is critical to avoid errors
• AI enrichment reduces the time an analyst spends interpreting raw log data
• Natural-language interfaces lower the barrier to entry for SOC operations

---

🔒 Security Notes

• API account used with read-only scope (principle of least privilege)
• Secrets managed via .env file — never committed to version control
• WAZUH_VERIFY_SSL=true recommended in production with valid certificates
• Indexer port 9200 should be restricted to trusted IPs only in production

---

🛠️ Skills Demonstrated

Technical Skills

• SIEM deployment and configuration (Wazuh)
• Python async development (httpx, MCP SDK)
• REST API integration (JWT authentication, OpenSearch queries)
• Multi-VM network configuration (VMware, bridged networking)
• AI tool integration (Claude Desktop, MCP protocol)
• Slack API (Incoming Webhooks, Block Kit formatting)

Security Skills

• Alert triage and severity classification
• Vulnerability management workflow
• SOC operations and incident analysis
• Security report generation

---

⚙️ Installation

Prerequisites

• Python 3.10+
• Wazuh Manager + Indexer (Ubuntu Server)
• Claude Desktop
• Slack Incoming Webhook URL

Setup

git clone https://github.com/cisse-lalya/wazuh-mcp-soc.git
cd wazuh-mcp-soc
pip install -r requirements.txt
copy .env.example .env
notepad .env   # Fill in your Wazuh + Slack credentials

.env Configuration

WAZUH_API_URL=https://<WAZUH_VM_IP>:55000
WAZUH_API_USER=wazuh
WAZUH_API_PASSWORD=your_password

WAZUH_INDEXER_URL=https://<WAZUH_VM_IP>:9200
WAZUH_INDEXER_USER=admin
WAZUH_INDEXER_PASSWORD=your_password

WAZUH_VERIFY_SSL=false
SLACK_WEBHOOK_URL=https://hooks.slack.com/services/XXX/YYY/ZZZ

Test Connection

python -c "from dotenv import load_dotenv; load_dotenv(); import asyncio; from wazuh_client import WazuhClient; c=WazuhClient(); print(asyncio.run(c.get_agents_status()))"

---

🧰 Tech Stack

Wazuh 4.11.2 · OpenSearch · Claude Desktop · MCP Protocol · Python 3.13 · httpx · Slack Webhooks · VMware · Ubuntu Server · Windows 10 · Kali Linux

---

👤 About Me

Cybersecurity student and Junior SOC Analyst, passionate about bridging network operations and security through automation and AI-driven tooling. This project reflects my approach to security: building practical, integrated solutions rather than studying tools in isolation.

Connect with me:

• GitHub: @cisse-lalya
• LinkedIn: linkedin.com/in/ndeyelalyacisse