Panguard AI

Panguard AI

Panguard AI by Panguard provides AI agent security for developers, scanning MCP skills for prompt injection, tool poisoning, and supply chain threats. It offers a one-command install, 747 detection ru

AI securityMCP scanningprompt injection detectionsupply chain protectiondeveloper tools
Open site
image for Panguard AIOpen site

What is Panguard AI?

Panguard AI is an open-source security tool for AI agents that scans MCP skills for prompt injection, tool poisoning, and supply chain threats. It audits every skill before it runs, catches known threats with community ATR (Agent Threat Rules), and uses AI analysis to catch unknown threats. The tool installs in one command, runs offline with zero telemetry, and is MIT-licensed. It automatically detects popular AI agents like Claude Code, Cursor, Windsurf, and VS Code Copilot.

Application scenarios

  • Agent security auditing

    Scan any MCP skill before running it to catch prompt injection and tool poisoning.

  • Runtime agent protection

    Guard each agent action at runtime after scanning skills and MCP servers.

  • Community threat sharing

    Contribute and receive new ATR rules to protect the entire ecosystem from emerging attacks.

  • Offline security scanning

    Run the tool without an internet connection or sign-up, making it suitable for air-gapped environments.

  • Supply chain risk detection

    Identify CRITICAL or HIGH security risks in MCP skills—1.9% of 67,799 scanned skills had such risks.

  • CVE vulnerability monitoring

    Stay protected against real CVEs like MCPJam (CVSS 9.8), Claude Code exploits (CVSS 8.7), and Azure MCP Server SSRF (CVSS 7.5).

Core Features

  • One-command install

    Run `curl -fsSL https://get.panguard.ai | bash` to install in 30 seconds with no signup required.

  • 768 ATR rules

    Community-powered Agent Threat Rules that accumulate faster than attacks multiply, with a 97.2% Garak recall rate.

  • AI-based unknown threat detection

    Beyond known rules, AI analysis catches novel attacks that don't match existing signatures.

  • Auto-detection of AI agents

    Automatically identifies and works with Claude Code, Claude Desktop, Cursor, Hermes Agent, Codex CLI, Windsurf, VS Code Copilot, and many more.

  • Self-hosted with zero telemetry

    Runs entirely on your infrastructure with no data sent externally, ensuring privacy.

  • Runtime guard

    After scanning, it monitors each agent action during execution to block threats in real time.

  • MIT-licensed open source

    Free forever, with the ability to inspect, modify, and contribute to the codebase.

  • Merged upstream rules

    Integrates detection rules from Cisco and Microsoft for broader coverage.

Target users

Developers and security engineers building or deploying AI agents. Teams using Claude Code, Cursor, Windsurf, VS Code Copilot, Gemini CLI, or any of the 20+ supported agents. Open-source contributors who want to share and benefit from community-driven threat detection. Organizations that need offline, self-hosted security for agentic workflows without vendor lock-in.

How to use Panguard AI?

Install with a single terminal command: curl -fsSL https://get.panguard.ai | bash. The tool auto-detects your AI agents and begins scanning MCP skills immediately. No signup, account, or configuration required—it runs offline and guards each agent action at runtime. For detailed usage, refer to the official documentation at https://panguard.ai.

Pricing and free trial

Panguard AI is free and open source under the MIT license. The website states "MIT licensed. Open source. Free · Open source · MIT" and "free forever." No paid tiers or trial periods are mentioned—the entire tool is available at no cost.

Effect review

Panguard AI addresses a genuine gap in AI agent security: the lack of shared standards for prompt injection and tool poisoning. With 768 ATR rules, a 97.2% Garak recall rate, and real CVE coverage (including a CVSS 9.8 MCPJam vulnerability), it provides credible protection out of the box. The one-command install and zero-telemetry design make it practical for developers who want security without friction. However, the tool's effectiveness ultimately depends on community participation—the faster attacks multiply, the more critical it is for teams to contribute new rules. For teams already running AI agents, this is a low-risk, high-value addition to their security stack.

Frequently Asked Questions

What is Panguard AI?
Panguard AI is an AI agent security tool that scans MCP skills for prompt injection, tool poisoning, and supply chain threats, offering a one-command install and 747 detection rules.
How do I install Panguard AI?
You can install Panguard AI with a single command, making it quick and easy for developers to set up.
What threats does Panguard AI detect?
It detects prompt injection, tool poisoning, and supply chain threats in MCP skills.
How many detection rules does Panguard AI have?
Panguard AI includes 747 detection rules to identify security vulnerabilities.
Who is Panguard AI designed for?
It is designed for developers who build with AI agents and need to secure their MCP skills.
Is Panguard AI free to use?
Pricing details are not provided; please check the official Panguard AI website for current pricing information.

Panguard AI - AI Tool Detail

Panguard AI by Panguard provides AI agent security for developers, scanning MCP skills for prompt injection, tool poisoning, and supply chain threats. It offers a one-command install, 747 detection ru

Category：AI Plugin/Extension

Visit Link：https://panguard.ai/

Tags：AI security、MCP scanning、prompt injection detection、supply chain protection、developer tools