MCP正在迅速被采用。安全指导落后。这份清单为安全工程师、平台团队和技术领导者提供了一个清晰且可操作的基准，以保障MCP的安全......
概览
一份实用的、由社区维护的安全清单，适用于构建和部署服务器及人工智能代理基础设施的团队。 MCP正在迅速被采用。安全指导滞后。 这份清单为安全工程师、平台团队和技术领导者提供了保障MCP部署的保障——无论你是发布内部工具还是面向客户的AI代理。 它是供应商特定的、完整的，或者作为完整安全审查的替代品。它是一个起点。 该清单提供了JSON和YAML版本，可用于集成到CI/CD管道、合规工具或自定义仪表盘中： 请以更友好的方式浏览清单： 该清单通过社区的反馈不断完善。如果您在实际MCP部署中遇到漏洞、配置错误或攻击模式——我们非常欢迎分享。 请参阅 CONTRIBUTING.md 了解如何参与。
概述（AI 翻译）
一份实用的、由社区维护的安全检查表，适用于构建和部署 **模型上下文协议（MCP）**服务器和人工智能代理基础设施的团队。
为什么会有这样的存在
MCP正在迅速被采用。安全指导滞后。
这份清单为安全工程师、平台团队和技术领导者提供了一个清晰且可操作的基线，用于保障MCP部署——无论您是发布内部工具还是面向客户的AI代理。
它**不是*厂商特定、完整或替代完整安全审查的。它只是一个起点。
📋 清单
| 清单 | 受众 | 描述 |
|---|---|---|
| 认证与授权 | 全部 | 身份、令牌范围与访问控制 |
| 输入验证与提示注入 | 工程师 | 工具执行前的输入净化 |
| 工具与资源暴露 | 工程师 / 架构师 | 限制MCP工具的爆炸半径 |
| API 会话安全 | 平台团队 | 从代理手中保护入站会话 |
| 监控与可观察性 | 安全运营 | 要记录、提醒和审查的内容 |
| 网络与基础设施 | 平台团队 | 网络层加固 |
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
🔐 MCP Security Checklist
A practical, community-maintained security checklist for teams building and deploying Model Context Protocol (MCP) servers and AI agent infrastructure.
Why This Exists
MCP is being adopted rapidly. Security guidance is lagging behind.
This checklist gives security engineers, platform teams, and technical leaders a clear, actionable baseline for securing MCP deployments — whether you’re shipping an internal tool or a customer-facing AI agent.
It is not vendor-specific, complete, or a replacement for a full security review. It is a starting point.
📋 The Checklists
| Checklist | Audience | Description |
|---|---|---|
| Authentication & Authorization | All | Identity, token scope, and access control |
| Input Validation & Prompt Injection | Engineers | Sanitizing inputs before tool execution |
| Tool & Resource Exposure | Engineers / Architects | Limiting blast radius of MCP tools |
| API Session Security | Platform Teams | Securing inbound sessions from agents |
| Monitoring & Observability | SecOps | What to log, alert on, and review |
| Network & Infrastructure | Platform Teams | Network-layer hardening |
| CISO Summary | CISOs / Leadership | Non-technical risk summary |
✅ Quick-Start: Top 10 Controls
If you do nothing else, cover these:
- Never expose MCP over the public internet without mTLS or equivalent.
- Scope every tool to the minimum necessary permissions.
- Validate and sanitize all inputs before they reach tool execution.
- Log every tool invocation with the originating session context.
- Set rate limits on both the MCP server and any downstream APIs it calls.
- Treat agent sessions as untrusted by default — validate intent, not just auth tokens.
- Separate read and write tool categories; require explicit approval for write operations in sensitive contexts.
- Rotate credentials used by MCP servers on a defined schedule.
- Monitor for behavioral anomalies: unusual tool chains, high-frequency calls, off-hours access.
- Conduct a tool inventory review before every production deployment.
🗂️ Machine-Readable Version
A JSON and YAML version of the checklist is available for integration into CI/CD pipelines, compliance tooling, or custom dashboards:
安装
This server does not publish a one-line install command.
Open the repository installation guide 新手？先看 mcpskill.org 教程 →