Skip to content

README_EN.md

Latest commit

 

History

History
101 lines (69 loc) · 4.84 KB

README_EN.md

File metadata and controls

101 lines (69 loc) · 4.84 KB
Raw

AutoRedTeam-Orchestrator

A local-first, MCP-native security automation workbench for authorized testing and AI/MCP attack-surface review.

Python License Version Status

中文 · English · Русский

Capability Maturity · Security Model · Security Audits

Beta / Research Preview — prioritizes static analysis, dry-runs, and local labs. Exploit, lateral movement, persistence, C2, exfiltration, and other high-risk capabilities are restricted experimental features; policy and sandbox metadata are not OS- or container-level isolation.

Overview

AutoRedTeam-Orchestrator provides composable security capabilities in a single codebase, exposed through three entry points: an MCP server, a Python SDK, and a Typer CLI. The primary line is AI-assisted security automation — letting AI editors and agents use controlled security capabilities over MCP, and statically auditing the attack surface of AI/MCP systems themselves.

It fits authorized labs, security-automation development, AI/MCP security review, CTFs, and teaching. It is not a turnkey enterprise platform, an autonomous attacker, or a production C2.

Capabilities

Domain Status Scope
Recon and vulnerability detection (JSON/SARIF) Beta Authorized targets only; no public accuracy benchmark yet
Static AI/MCP surface self-audit Preview Read-only AST → SARIF; covers FastMCP and low-level SDK
CVE intelligence and PoC Preview NVD sync, Nuclei-compatible execution
Reporting (JSON/SARIF/HTML) Beta HTML escaping hardening in progress
Orchestration / exploit / post-exploit / C2 / lateral / persistence Restricted Experimental Disposable isolated labs only

See Capability Maturity for full definitions.

Quick start

git clone https://github.com/Coff0xc/AutoRedTeam-Orchestrator.git
cd AutoRedTeam-Orchestrator
pip install -r requirements.txt
python -m cli.main --help

Read-only, no-network commands to start:

python -m cli.main ai-surface scan --path handlers --format sarif -o surface.sarif
python -m cli.main capabilities manifest --profile safe

AI/MCP self-audit

Run a purely static audit of your own repository: inventory the attack surface of MCP server and AI agent tools, with results delivered at file:line precision to GitHub Code Scanning. No target, network, secret, or authorization required.

Run as a GitHub Action on every PR (full example in self-audit.example.yml):

- uses: Coff0xc/AutoRedTeam-Orchestrator@v3.1
  with:
    mode: self-audit
    path: '.'
    severity-threshold: high

Or locally:

python -m cli.main ai-surface scan --path . --format sarif        # MCP handler tool surface
python -m cli.main ai-surface scan-mcp-config --path .mcp.json    # broad commands and plaintext secrets
python -m cli.main ai-surface scan-skills --path ./skills         # high-risk instruction markers

Add --auth-mode lenient to drop project-specific authorization findings when auditing external repositories.

Entry points

Entry point Path Position
MCP Server mcp_stdio_server.py Fail-closed safe profile by default; trusted-local stdio only
Python SDK autort/ From a source checkout
Typer CLI cli/main.py Primary local-analysis and development entry

Authorized scans generate network traffic — use them only against localhost, a disposable lab, or a target under written authorization:

python -m cli.main scan http://127.0.0.1:8000 --full
python -m cli.main detect http://127.0.0.1:8000 -c sqli,xss,ssrf

Security and scope

  • MCP capability profiles (safe/scan/active-lab/full) filter tools at registration but do not replace authentication, target scope, independent approval, or an isolated executor.
  • High-risk capabilities default to dry-run and should be enabled only in disposable, isolated environments.
  • Authentication is attached per tool decorator and is not yet a uniform boundary across every registered surface.

See Security Model and Capability Maturity.

License and disclaimer

MIT License — see LICENSE.

For explicitly authorized security testing, internal validation, education, and local labs only. Users must follow applicable law and obtain written authorization from the target owner. Unauthorized scanning, exploitation, persistence, exfiltration, or evasion of security controls is prohibited.