Autonomous OODA-loop pentesting agent. Hermes brain, Kali + Playwright tools over MCP, ATT&CK as the Orient language.
Python ai-agentsautonomous-agentskali-linuxllm-agentsoffensive-securityooda-looppentestingplaywrightsecurity
# config.yaml.example — documented variant with placeholders.
# Copy to the profile's config.yaml and adjust if needed after validating live.
# MCP syntax verified against:
# https://github.com/NousResearch/hermes-agent/blob/main/website/docs/user-guide/features/mcp.md
model:
default: "mistralai/mistral-large-4-0" # OpenRouter model prepared for authorized pentesting
mcp_servers:
kali:
url: "http://kali:8900" # Kali MCP (container `kali`). NO headers/token.
tools:
# Whitelist of the real kali-mcp server tools (name as exposed by the server).
include: [kali_run, kali_exec, iam_enum, iam_simulate, get_scan_status, read_tool_logs, prowler_scan, record_finding, read_findings, server_info]
prompts: false
resources: false
playwright:
url: "http://playwright-mcp:8931/mcp" # Browser MCP (headless Chromium) for the web part.
tools:
prompts: false # NO include: the image's browser tools are exposed.
resources: false
# Two toolsets: both MCPs (Kali + browser). ZERO native tools (native browser/web disabled).
toolsets:
- kali
- playwright
terminal:
backend: "local" # Hermes does not use a native terminal; it runs on the Kali via kali_run.