MCPSECBENCH
收藏资源简介:
MCPSECBENCH是一个全面的安全基准和测试环境,用于评估模型上下文协议(MCP)的安全性。该数据集由17种攻击类型组成,涵盖了MCP的四个主要攻击面:用户交互、MCP客户端、MCP传输和MCP服务器。数据集包括丰富的提示数据集、示例MCP客户端、多个易受攻击和恶意服务器以及用于传输层攻击的脚本。研究人员可以在该环境中灵活地评估其MCP主机、客户端、服务器和传输协议的安全性,并轻松扩展以支持新的攻击类型。
MCPSECBENCH is a comprehensive security benchmark and test environment for evaluating the security of Model Context Protocol (MCP). The dataset consists of 17 attack types, covering four primary attack surfaces of MCP: user interaction, MCP client, MCP transport, and MCP server. It includes a rich set of prompt datasets, sample MCP clients, multiple vulnerable and malicious servers, as well as scripts for transport-layer attacks. Researchers can flexibly evaluate the security of their MCP hosts, clients, servers, and transport protocols in this environment, and easily extend it to support new attack types.




