# Agentic Security — LLM/MCP Red-Team Scanner

> Agentic Security is a Python tool to probe LLM apps with attack prompts and run scans; it also ships an MCP server entrypoint for tool-based workflows.

## Install

Save as a script file and run:

## Quick Use

1. Install:
   ```bash
   pip install agentic_security
   ```
2. Start the scanner:
   ```bash
   agentic_security --help
   agentic_security
   ```
3. (Optional) Install as an MCP server (from a clone):
   ```bash
   pip install -U mcp
   mcp install agentic_security/mcp/main.py
   ```

## Intro

Agentic Security is a Python tool to probe LLM apps with attack prompts and run scans; it also ships an MCP server entrypoint for tool-based workflows.

- **Best for:** security teams validating LLM safety, jailbreak resistance, and tool-call hardening
- **Works with:** Python, local datasets (CSV), optional MCP server install
- **Setup time:** 15–45 minutes

## Practical Notes

- Data point: the README example shows the server running on port 8718.
- Quant: run scans against staging first; promote to CI only after false-positive tuning.

## Pattern: build a reproducible red-team harness

To avoid one-off, non-repeatable results:
- pin the dataset version (CSV),
- log model + parameters,
- and store scan outputs as artifacts.

## Pattern: scan tool servers too

If your agent uses MCP servers, treat them as part of your attack surface:
- scan repo code,
- scan remote URLs before deployment,
- and add allowlists for file/network capabilities.

### FAQ

**Q: Is it open-source?**
A: Yes. The repo is Apache-2.0 licensed.

**Q: Can it run without an MCP client?**
A: Yes. You can run the scanner directly as a CLI.

**Q: What’s the safest rollout?**
A: Start on staging and log everything; then gradually enable CI gates.

## Source & Thanks

> Source: https://github.com/msoedov/agentic_security
> License: Apache-2.0
> GitHub stars: 1,867 · forks: 249

---

<!-- ZH -->

## 快速使用

1. 安装：
   ```bash
   pip install agentic_security
   ```
2. 启动扫描器：
   ```bash
   agentic_security --help
   agentic_security
   ```
3.（可选）作为 MCP server 安装（从仓库克隆目录执行）：
   ```bash
   pip install -U mcp
   mcp install agentic_security/mcp/main.py
   ```

## 简介

Agentic Security 是用于扫描与压测 LLM 应用的 Python 工具：可用攻击提示词做探测与基准化复现，并提供 MCP server 入口，便于接入工具化工作流、CI 与安全评审流程。

- **适合谁：** 想评估 jailbreak 风险、工具调用安全性与防护强度的安全团队
- **可搭配：** Python、本地 CSV 数据集、可选的 MCP server 安装方式
- **准备时间：** 15–45 分钟

## 实战建议

- 数据点：README 示例里服务跑在 8718 端口。
- 量化建议：先对 staging 跑，调好误报率后再接入 CI。

## 常用打法：把红队评测做成可复现的 harness

避免一次性“跑完就算”的评测：
- 固定数据集版本（CSV），
- 记录模型与参数，
- 把结果作为工件保存。

## 常用打法：把 MCP server 也纳入扫描面

如果 agent 使用 MCP server，就把它当攻击面的一部分：
- 扫代码，
- 部署前扫远程 URL，
- 对文件/网络能力做 allowlist。

### FAQ

**开源吗？**
答：是的，仓库为 Apache-2.0。

**不接 MCP 也能用吗？**
答：能。它本身就可以作为 CLI 运行。

**最稳的上线方式？**
答：先在 staging 跑并记录日志，再逐步接入 CI 门禁。

## 来源与感谢

> Source: https://github.com/msoedov/agentic_security
> License: Apache-2.0
> GitHub stars: 1,867 · forks: 249


---
Source: https://tokrepo.com/en/workflows/agentic-security-llm-mcp-red-team-scanner
Author: Agent Toolkit