Cybersecurity-MCP-Server

Cybersecurity-MCP-Server (AynOps) is a high-performance open-source Model Context Protocol (MCP) server designed to provide AI agents with real-time cybersecurity reconnaissance, attack surface mapping, and automated OSINT telemetry capabilities.

Purpose

Modern large language models run in sandboxed, non-deterministic execution environments and lack native raw socket access, packet construction, and real-time network reconnaissance primitives. When asked to audit an attack surface, AI agents are constrained by static training cutoff boundaries and cannot query live network interfaces. AynOps bridges this operational gap: built on the MCP specification and using the FastMCP framework, it exposes an extensible set of deterministic cybersecurity reconnaissance tools to LLM agents (such as Cursor, VS Code, Claude Desktop, and custom MCP clients) over standard JSON-RPC 2.0 stdio transport. The project targets cybersecurity engineers, red and blue teams, and security researchers, providing deterministic automation capabilities.

Key Capabilities

• 100% local-first execution: self-hosted local daemon, zero telemetry leakage, protecting target confidentiality.
• 3-wave asynchronous reconnaissance engine: parallelized execution pipeline that can reduce end-to-end full reconnaissance time by up to 70%.
• Structured heuristic signal extraction: converts raw terminal output into normalized, machine-readable threat telemetry data.
• Comprehensive boundary mapping: authoritative WHOIS (RFC 3912), comprehensive DNS RR resolution, multi-threaded dictionary-based subdomain brute-forcing.
• Certificate Transparency log extraction: real-time queries to the crt.sh append-only Merkle tree for passive subdomain and wildcard SSL discovery.
• Dangling DNS and subdomain takeover detection: signature resolution of CNAME pointers for 8+ cloud providers (AWS S3, Azure, GitHub Pages, Heroku, Shopify, Fastly, Ghost, Pantheon), with active HTTP verification probes.
• Multi-cloud bucket auditing: generates domain keyword permutations and checks for unauthenticated public access in AWS S3, Azure Blob Storage, and Google Cloud Storage.
• RFC-compliant email security assessment: audits SPF (v=spf1), DKIM selectors, and DMARC enforcement policies (p=reject / quarantine / none), and provides a quantitative score (0–100).
• Multi-layer web technology fingerprinting: a 4-layer non-intrusive technology identification engine—HTTP response headers (Server, X-Powered-By, CDN markers), Set-Cookie cookie names, <meta name="generator"> tag parsing, HTML/resource path substring matching—returning structured detection results with confidence scores (0–100) and evidence strings.
• Deep HTTP redirect chain and TLS downgrade tracing: hop-by-hop 3xx status analysis, detecting SSL stripping, RFC 1918 internal IP address leakage, redirect loops, and cross-domain jumps.
• Real-time NIST NVD v2 CVE lookup: direct integration with the NIST National Vulnerability Database REST API to retrieve CVSS v3.1/v2 base scores, CWE vectors, and exploit references, with no API key required.
• Cognitive threat analysis meta-prompt: a native MCP prompt template that transforms raw scan telemetry into a prioritized vulnerability matrix and remediation roadmap.
• Container-ready architecture: ships with a production-grade Dockerfile preinstalled with Nmap and Python 3.12, enabling zero-dependency stdio deployment.

Usage

Prerequisites

• Python runtime: Python >= 3.12
• Network scanner: Nmap binary installed and accessible in the system PATH
• Environment and package manager: uv (recommended) or pip
• Version control: Git
• MCP client: Cursor, VS Code (MCP extension), Claude Desktop, Continue.dev, Zed, or any client implementing the MCP JSON-RPC protocol
• Container runtime (optional): Docker Engine / Docker Desktop

Installation

Option 1: Local virtual environment setup (recommended). Use uv or the standard Python toolchain. Nmap must be installed locally and registered in the system PATH (on Windows, add C:\Program Files (x86)\Nmap or C:\Program Files\Nmap to Path; on macOS, Homebrew can be used; on Linux, use the system package manager). Then insert the AynOps server configuration into the client MCP configuration file, always providing the .venv Python binary and the absolute path to server.py. ABUSEIPDB_API_KEY and HIBP_API_KEY are optional and are only needed for ip_reputation and hibp_check; the remaining tools run with zero API keys. Option 2: Containerized deployment via Docker. AynOps ships with a production-grade Dockerfile preinstalled with Python 3.12, uv, and nmap. The container must run in interactive mode (-i), and the detached flag (-d) must not be used, in order to maintain the stdio communication pipe required by the MCP protocol.

Tools and Prompts

AynOps provides 17 specialized reconnaissance tools and structured MCP prompt templates. The full_recon tool orchestrates 3 waves of concurrent pipelines, minimizing scan latency while preserving dependency order between tools. It supports natural language tool invocation, port scan configurations (basic, service, etc.), and the threat_analysis prompt for end-to-end threat modeling. Typical scenarios include: CVE patch verification and vulnerability correlation, dangling DNS subdomain takeover classification, unauthenticated multi-cloud storage exposure, and HTTP redirect chain and TLS downgrade analysis.

Use Cases

• Attack surface mapping in authorized penetration testing and red team operations
• Blue team defensive posture auditing and vulnerability management
• CVE patch verification and vulnerability correlation
• Subdomain takeover investigation and cloud storage exposure checks
• Email security (SPF/DKIM/DMARC) compliance assessment

Limitations

• OS detection privileges: Nmap OS fingerprinting (scan_type="os") relies on raw TCP/IP socket construction (-O) and requires elevated root/Administrator privileges on the host system.
• Upstream OSINT rate limits: tools that access third-party endpoints (such as crt.sh, NIST NVD v2 REST API) are subject to upstream service availability and rate limits.
• API key dependency: ip_reputation and hibp_check require valid API keys configured in environment variables to return live telemetry.
• Container stdio lifecycle: when deployed via Docker, the container must remain attInteractive stdio stream added; running in detached daemon mode (-d) will terminate the MCP pipeline.
• LLM client behavior: Tool call decisions, parameter synthesis, and prompt chaining are controlled by the connected LLM client; security filters of some LLM providers may reject certain queries (e.g., "Port scan example.com"), so it is recommended to phrase prompts defensively.
• Execution timeout: Large operations (e.g., a full 65535-port scan) may exceed the client's default JSON-RPC timeout threshold (typically 60–120 seconds).

Legal and Ethical

Perform active network scanning and vulnerability assessment only against infrastructure you own or have explicit written authorization to test. Passive OSINT tools operate only against publicly announced records and standard web responses; active scanning tools send active network traffic, and unauthorized scanning may violate the CFAA, GDPR, and local network regulations. For test demonstrations, you may use the officially designated Nmap test host scanme.nmap.org.