📄 中文摘要
模型上下文协议（MCP）正在迅速发展，数千名开发者在短时间内发布了MCP服务器，使AI代理能够访问数据库、文件系统、API和内部工具。然而，最近的研究发现，超过8000个MCP服务器在互联网上公开暴露且没有身份验证，任何AI代理或攻击者都可以连接到这些服务器，自由列出其工具并调用它们。这种情况带来了重大的安全隐患，开发者需要立即检查和审计自己的服务器，以防止潜在的安全灾难。
MCP 服务器安全：大多数开发者忽视的风险
出处: MCP Server Security: The Risks Most Developers Are Ignoring
发布: 2026年3月14日
📄 English Summary
MCP Server Security: The Risks Most Developers Are Ignoring
The Model Context Protocol (MCP) is rapidly evolving, with thousands of developers deploying MCP servers that allow AI agents to access databases, filesystems, APIs, and internal tools. However, recent research has uncovered over 8,000 MCP servers publicly exposed on the internet without authentication, enabling any AI agent or attacker to connect, enumerate their tools, and invoke them freely. This situation poses significant security risks, necessitating developers to check and audit their servers immediately to prevent potential security disasters.