Wazuh-MCP-Server MCP — Cas d'usage, Installer & Démo en direct

Pourquoi l'utiliser

Fonctionnalités clés

48 tools across alerts, agents, vulnerabilities, active response, compliance
Output sanitization — redacts credentials before they reach the LLM
Rate limiting + circuit breakers to protect Wazuh Manager
RBAC per tool; audit logs for destructive actions
Deploys via Docker Compose; supports Wazuh 4.8–4.14

Démo en direct

Aperçu en pratique

wazuh.replay ▶ prêt

0/0

Installer

Choisissez votre client

~/Library/Application Support/Claude/claude_desktop_config.json · Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "wazuh": {
      "command": "uvx",
      "args": [
        "Wazuh-MCP-Server"
      ],
      "_inferred": true
    }
  }
}

Ouvrez Claude Desktop → Settings → Developer → Edit Config. Redémarrez après avoir enregistré.

~/.cursor/mcp.json · .cursor/mcp.json

{
  "mcpServers": {
    "wazuh": {
      "command": "uvx",
      "args": [
        "Wazuh-MCP-Server"
      ],
      "_inferred": true
    }
  }
}

Cursor utilise le même schéma mcpServers que Claude Desktop. La config projet l'emporte sur la globale.

VS Code → Cline → MCP Servers → Edit

{
  "mcpServers": {
    "wazuh": {
      "command": "uvx",
      "args": [
        "Wazuh-MCP-Server"
      ],
      "_inferred": true
    }
  }
}

Cliquez sur l'icône MCP Servers dans la barre latérale Cline, puis "Edit Configuration".

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "wazuh": {
      "command": "uvx",
      "args": [
        "Wazuh-MCP-Server"
      ],
      "_inferred": true
    }
  }
}

Même format que Claude Desktop. Redémarrez Windsurf pour appliquer.

~/.continue/config.json

{
  "mcpServers": [
    {
      "name": "wazuh",
      "command": "uvx",
      "args": [
        "Wazuh-MCP-Server"
      ]
    }
  ]
}

Continue utilise un tableau d'objets serveur plutôt qu'une map.

~/.config/zed/settings.json

{
  "context_servers": {
    "wazuh": {
      "command": {
        "path": "uvx",
        "args": [
          "Wazuh-MCP-Server"
        ]
      }
    }
  }
}

Ajoutez dans context_servers. Zed recharge à chaud à la sauvegarde.

claude mcp add wazuh -- uvx Wazuh-MCP-Server

Une seule ligne. Vérifiez avec claude mcp list. Supprimez avec claude mcp remove.

Cas d'usage

Usages concrets : Wazuh-MCP-Server

Triage overnight alerts in 10 minutes instead of 2 hours

👤 SOC analysts ⏱ ~20 min advanced

Quand l'utiliser : Shift start. Hundreds of alerts overnight. You need to find the real ones without reading every row.

Prérequis

Wazuh Manager API creds — Wazuh UI > API credentials

Déroulement

Summarize

For last 12 hours: group alerts by rule group, count severity 10+ per group, top 5 agents with most high-sev alerts.✓ Copié

→ Ranked summary
Investigate

For top agent, pull the 10 most recent high-sev alerts with full details.✓ Copié

→ Detailed events
Determine action

Based on these events, is this a true positive? If yes, propose response: isolate agent / disable account / create ticket.✓ Copié

→ Verdict + action plan

Résultat : Faster MTTD/MTTR without more eyeballs.

Pièges

Auto-triggering active response before confirming — Keep active response tools behind a confirmation gate

Weekly vulnerability posture report

👤 Security managers ⏱ ~30 min intermediate

Quand l'utiliser : Friday report for leadership. How many CVEs open? Where? Trend vs last week?

Déroulement

Pull vulnerabilities

From Wazuh vulnerability module: total open CVEs by severity, top 10 agents by critical CVE count.✓ Copié

→ Numbers + top agents
Compare to last week

Compare counts to last Friday's report [paste]. What improved, what regressed?✓ Copié

→ Delta analysis
Prioritize

List the 5 critical CVEs affecting most hosts that should be patched this weekend.✓ Copié

→ Actionable patch list

Résultat : Board-ready security posture summary.

Pièges

Vuln data only as fresh as last scan — Check scan timestamps before drawing conclusions

Combiner avec : notion

Isolate a compromised host on confirmation

👤 SOC on-call ⏱ ~10 min advanced

Quand l'utiliser : Alert fires: unambiguous malware on endpoint. You need to isolate fast.

Déroulement

Confirm scope

For agent 'ws-4412': show active processes with parent PID, recent listening ports, and last 20 alerts. Is this clearly compromised?✓ Copié

→ Risk assessment
Isolate

Isolate agent ws-4412 via active response. Require my explicit confirm first.✓ Copié

→ Confirmation prompt, then isolation + audit log entry

Résultat : Compromised host isolated within minutes.

Pièges

False-positive isolation of prod server — Double-check agent ID and classification before confirming; keep an out-of-band way to re-enable

Combinaisons

Associez-le à d'autres MCPs pour un effet X10

wazuh + notion

Weekly security digest to exec

Compose a weekly Wazuh summary and create a Notion page in 'Security Weekly'.✓ Copié

Outils

Ce que ce MCP expose

Outil	Entrées	Quand appeler	Coût
list_alerts	filter: obj, from, to, limit	Investigate by time/severity/rule	1 API call
get_agent	agent_id	Profile a specific endpoint	1 API call
list_vulnerabilities	agent_id?, severity?	Vuln posture checks	1 API call
trigger_active_response	agent_id, command, arguments?	Isolate / kill process / disable account — destructive; confirm	1 API call + physical effect
compliance_report	framework: 'pci'\|'hipaa'\|'nist', scope	Audit readiness	1 API call

Coût et limites

Coût d'exécution

Quota d'API: Bounded by Wazuh Manager API limits
Tokens par appel: Alert listings can be huge — always filter by time + severity
Monétaire: Free. Wazuh itself is open-source.
Astuce: Use limit aggressively. Don't list a week's alerts without grouping.

Sécurité

Permissions, secrets, portée

Portées minimales : Wazuh API user scoped read-only for inspection; elevate for active response

Stockage des identifiants : Wazuh API creds in env or Docker secrets

Sortie de données : Alert/event data goes to your LLM provider

Ne jamais accorder : Active-response scope to a dev/eval instance

Active response tools cause real actions on endpoints. Gate behind human confirmation.
Alert payloads may contain sensitive info (filenames, users); consider Presidio-style redaction upstream.