Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@blackarch-AI MCP Server scan the target 10.0.0.1 with nmap"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
A scope-gated MCP server and set of Claude Code Skills for red-team and blue-team security workflows on BlackArch Linux.
MCP server (src/blackarch_ai_mcp/) — wraps local security tools (nmap, gobuster, nikto, hydra, plus local blue-team audits) as MCP tools, so any MCP-compatible client can call them.
Claude Code Skills (.claude/skills/) — markdown playbooks that walk through red-team and blue-team workflows using those tools.
Related MCP server: Pentest MCP
No red-team tool will run against a target that isn't explicitly authorized. Every tool that touches a network target (nmap_scan, gobuster_dir, nikto_scan, hydra_bruteforce) checks scope.yaml first and refuses with a clear message if the target isn't listed with valid, non-expired authorization.
cp scope.example.yaml scope.yaml
# edit scope.yaml: add host/cidr, authorized_by, authorization_ref, expiresscope.yaml is gitignored — it's local, never published. See docs/SCOPE.md for the full contract.
Blue-team tools (audit_listening_ports, audit_suid_world_writable, review_auth_log, lint_ssh_config, audit_pacman_packages, crack_hash_offline) are local-only and not gated — they never touch a third party.
This project does not grant authorization to test anything. You are responsible for having explicit, written permission before pointing any red-team tool at a target. Unauthorized scanning/exploitation of systems you don't own or have written permission to test is illegal in most jurisdictions.
sudo pacman -S --needed python-uv
uv syncuv run python -m blackarch_ai_mcp.serverclaude mcp add blackarch-ai -- uv --directory /path/to/blackarch-ai run python -m blackarch_ai_mcp.serverThen in a Claude Code session, invoke a Skill (e.g. "run a host audit", or /blue-team-host-audit if slash-invocation is enabled) or call an mcp__blackarch-ai__* tool directly.
See docs/ARCHITECTURE.md for how the pieces fit together and docs/TOOLS.md for the full tool reference.
uv run pytestMIT — see LICENSE.
List listening ports (ss -tulnp) cross-referenced with firewall rules on this local host.
Inputs: no named parameters. Returns: result
Report orphan packages, available updates, and package integrity via pacman (read-only) on this local host.
Inputs: no named parameters. Returns: result
Find SUID and world-writable files under paths (default /usr /etc /opt /home) on this local host.
Inputs: paths. Returns: result
Check whether target is authorized in scope.yaml before running any red-team tool.
Inputs: target (required). Returns: result
Run an offline password-strength audit with hashcat/john against a hash file placed in data/hashes/.
Inputs: wordlist (required), hash_file (required), hash_mode (required), tool. Returns: result
Directory-brute-force url with gobuster. Refuses if the host is not in scope.yaml.
Inputs: url (required), wordlist, extensions. Returns: result
Credential brute-force target/service with hydra using local userlist/passlist files.
Requires confirm=True as an explicit double opt-in, and target must be in scope.yaml.
Inputs: target (required), service (required), passlist (required), userlist (required), +1 more. Returns: result
Check sshd_config at path against a baseline hardening ruleset on this local host.
Inputs: path. Returns: result
Inventory installed BlackArch security-tool packages, grouped by category (recon, exploitation, cracking, etc.).
Inputs: no named parameters. Returns: result
Run a nikto web vulnerability scan against target:port. Refuses if target not in scope.yaml.
Inputs: target (required), port. Returns: result
Run an nmap scan against target (profile: quick|full_tcp|service|udp). Refuses if target not in scope.yaml.
Inputs: target (required), profile. Returns: result
Summarize sshd auth log entries on this local host since since.
Inputs: since. Returns: result
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 12 tools
Most tools have clearly distinct purposes: scanning (nmap, nikto, gobuster), credential attacks (hydra), local auditing (audit_*, lint_*, review_*), and offline cracking (crack_hash_offline). However, check_scope is a utility tool that could be considered overlapping with the scope checks built into other tools, but it serves as a proactive check rather than a reactive refusal.
The naming pattern mixes styles: verbs like 'check_', 'nmap_scan', 'gobuster_dir', and 'hydra_bruteforce' use a tool_name format, while others like 'audit_listening_ports' and 'review_auth_log' use verb_noun. The category for local auditing tools is consistent, but the overall set lacks a single unified pattern.
12 tools is a reasonable count for a security auditing server covering reconnaissance, brute-forcing, local system auditing, and offline cracking. It's slightly high but each tool has a specific role, and no tool feels redundant. The scope is well-defined for a BlackArch-oriented toolset.
The toolset covers external scanning, web vulnerability scanning, brute-forcing, local host auditing, and password cracking. However, there are gaps: no network sniffing, exploitation, or privilege escalation tools, and no reporting or logging of results. For a comprehensive security assessment suite, these missing capabilities could hinder an agent's workflow.
MEOK MCP Hardening MCP — automated security red-team for any MCP server. Maps OWASP LLM Top 10
Offline methodology engine for authorized penetration testing, CTF, and security research.
Find, vet, and run MCP tools through a secure audited gateway with prompt-injection risk scoring
Secure Linux sandboxes for AI agents, as MCP tools.