A high-fidelity security intelligence service providing real-time auditing, static analysis, and semantic threat detection for the AI agent ecosystem.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install arc-security-mcp
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install arc-security-mcp using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is ARC Security MCP Server?
The ARC Security MCP Server is a specialized security intelligence platform designed to protect the AI-first ecosystem. Unlike traditional scanners built for human-readable code, this tool is engineered for agents to query safety metrics before executing or installing new components. It provides a robust defense layer for Openclaw Skills by leveraging a database built from over 743 manual audit findings across 361 skill surveys.
At its core, the server facilitates deep analysis of skill behavior, moving beyond simple regex patterns to Gen 2 intent analysis. This allows developers to detect sophisticated threats like capability-purpose mismatches, identity drift, and distributed attack chains. By integrating this server, users ensure that their Openclaw Skills implementations remain resilient against novel agentic attack vectors documented through 31 rounds of rigorous manual code audits.
ARC Security MCP Server Use Cases
- Verifying the safety status of a specific skill from the Openclaw Skills directory before deployment.
- Scanning skill source code for dangerous patterns such as shell injection, credential exposure, or self-modification.
- Detecting complex semantic threats and social engineering via LLM-powered intent analysis of skill documentation.
- Researching current ecosystem threat intelligence and documented agent attack classes like soul engineering or vishing.
How ARC Security MCP Server Works
- Connect your AI agent or MCP client to the server via the provided SSE endpoint to establish a secure intelligence link.
- Invoke the safety check tools to query the centralized database of known dangerous or malicious Openclaw Skills.
- Perform static analysis on the skill's source code to identify any of the 25 documented dangerous patterns.
- Utilize the AI-first intent analysis tool to evaluate the SKILL.MD file for capability-purpose mismatches and hidden exfiltration channels.
- Access the threat landscape intelligence to understand the broader context of findings, including critical and high-risk vulnerabilities across the ecosystem.
ARC Security MCP Server Setup
To integrate this security layer into your environment, point your MCP client configuration to the following SSE endpoint:
https://arcself.com/mcp/sse
No further local installation is required, as the server provides real-time security intelligence for your Openclaw Skills via the Model Context Protocol.
ARC Security MCP Server Data Schema & Taxonomy
The server organizes security data into several key structures to provide actionable intelligence for Openclaw Skills:
| Data Component | Description |
|---|---|
| Safety Registry | A database of 73+ known-dangerous skills verified through manual audits. |
| Pattern Library | 25 distinct dangerous pattern rules including regex, risk level, and mitigation steps. |
| Attack Taxonomy | Documentation of 22 agent-specific attack classes like agent-mediated kinetic action. |
| Audit Stats | Quantitative data on 743+ findings across critical and high severity levels. |
ARC Security MCP Server Advanced Features
- Generation 2 Intent Analysis that uses LLMs to catch social engineering attacks that no standard scanner can identify.
- Support for A2A (Agent-to-Agent) worm propagation detection and anti-safety training attack mitigation.
- Tailored security checklists specifically designed for different skill categories including financial, filesystem, and browser-based Openclaw Skills.
- Real-world intelligence derived from 31 manual audit rounds, providing data that goes beyond theoretical scanner output.
SKILL.md
Loading
Related Openclaw Skills
Security Audit
A unified security assessment tool that chains scanners and verifiers to harden and secure your entire agent skill ecosystem.
Metrics Dashboard
A lightweight operational monitoring tool to track agent health, API usage, and performance metrics through a text-based dashboard.
Memory Pruner
A specialized utility to automatically prune, compact, and manage the size of AI agent memory and log files.
Free Worker Dispatch
A smart routing utility that saves money by delegating routine content, research, and coding tasks to free AI models via OpenRouter.
Skill Differ
A specialized security utility to compare different versions of Openclaw Skills and detect new, potentially dangerous capabilities before updating.
Agent GitOps
Automated deployment, rollback, and version management for AI agent workflows and skills.
METADATA
Featured*
