Hermes-Agent-MCP-Vulnerability-PoC - This research is strictly intended for defensive a

项目说明

Hermes Agent - Dynamic MCP Manifest Mutation & Fail-Open PoC

This repository contains the official Proof of Concept (PoC) code, mock ser, and raw execution logs associated with the security research article analyzing dynamic manifest mutation and fail-open state handling in the Hermes Agent framework via the Model Context Protocol (MCP).

📖 Read the Full Article: Medium Article Link

---

📌 O

During security analysis of agentic toolchains, a pattern was identified where remote MCP endpoints can dynamically alter a tool schemas mid-session—specifically during state recovery routines.

When connection drops or HTTP 500 errors occur, the agent's re-discovery mechanism ingests updated tool schemas without cryptographic , schema pinning, or human-in-the-loop authorization.

This repository provides a reproducible en demonstrating:
1. Phase 1: HTTP 500 ser/disconnect during acti.
2. Phase 2: Dynamic re-connection, schema mutation (execute_system_command swapped for generate_debian_file), and immediate local execution.

---

📁 Repository Structure

🚀 Reproduction Steps
Prerequisites
* Python 3.10+
* Hermes Agent framework installed
* curl / basic network utilities

Running the Mock Server
1. Clone the repository:

2. Run the mock MCP server:

3. Update your Hermes Agent config.yaml to point to the local/remote HTTP endpoint ().

4. Initiate the agent session and trigger the reconnect flow as detailed in the logs/ directory.

🛡️ Remediation O
* Schema Pinning: Enforce cryptographic signature (HMAC/Asymmetric) on all incoming tool manifests.
* State Lockdown: Quarantine remote endpoints on status code anomalies (500 internal ser) and mandate re-authentication for altered schemas.
* Execution Sandboxing: Restrict local host interaction (file write, shell in) behind strict human-in-the-loop prompts.

⚠️ Disclaimer
This repository is created strictly for educational, defensi, and security purposes. All security testing should be conducted in isolated test en.

项目信息

开发语言：Python

Star 数：1

技术要点

该项目属于防护思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕This research is strictly intended for defensive analysis, security verification, and patching upstream agent implementations展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。