mcp-security-scan

当用户说『接了个MCP server不放心』『MCP工具能执行命令怕有风险』『怎么审计MCP权限』『第三方MCP会不会偷数据』，或要把某 MCP server(模型上下文协议)接进 agent、担心它是新攻击面时使用。把 MCP server 当『需审查的第三方』：扫工具清单里的 命令执行/文件系统写/网络外联/凭证暴露 四类风险，给风险评级与最小授权建议。理论根基：LGD 三律（有籍·有证·有门禁）。触发词：MCP安全、mcp security、MCP审计、MCP权限、第三方MCP、MCP风险、server扫描、协议安全。

凡自治之物 · LGD 治理学 · 创造术家族 ｜ 评测协议：skill-quality-gate

English

When to use

• When the user plans to connect a third-party MCP server to the current agent and needs to assess whether it introduces a new attack surface.
• When the user is concerned about MCP tools executing commands, file-write access or network egress.
• Whenever auditing MCP permissions and confirming LGD three-laws (registered · evidenced · gated) compliance — this skill must be used for pre-onboarding review.

Usage

Call the main script mcp_scan.py with the target MCP server's config or metadata as input. Specify target_server_name, access_config and risk concerns (e.g. command_exec, fs_write). The skill reads its built-in security rules, runs the simulated detection and returns a standardized JSON report with detailed risk categories and the LGD compliance verdict.

Workflow

First, receive the MCP server's identity and config description, confirming it is a "third party under review". Then run mcp_scan.py to score the four risk classes: command execution, filesystem write, network egress and credential exposure. Cross-validate against the LGD three-laws model that the permission boundary is closed and traceable. Finally, generate the report with risk rating (high/medium/low) and least-privilege recommendations, for the decision-maker to admit or circuit-break.

Evidence & safety statement

• Evidence: TBD (no evaluation / badge link yet; update here once it is produced)
• Version: 1.1.1
• Security: no independent security scan has been performed (TBD); the conclusion is pending and no endorsement is given until it is completed.

中文

何时用

当用户计划将第三方 MCP server 接入当前 Agent，需评估其是否构成新的安全攻击面时。 当检测到用户对 MCP 工具执行命令、文件写权限或网络外联存在风险疑虑时。 在涉及如何审计 MCP 权限及确认 LGD 三律（有籍·有证·有门禁）合规性时，必须使用此技能进行前置审查。

用法

调用主脚本 mcp_scan.py 并注入目标 MCP server 的配置信息或元数据作为输入参数。需指定 target_server_name、access_config 以及风险关注点（如 command_exec, fs_write）。技能将读取脚本内的安全规则，执行模拟检测逻辑，并返回标准化的 JSON 格式报告，包含详细的风险分类与 LGD 合规判定结果。

流程

首先接收 MCP server 的身份信息与配置描述，确认其作为“需审查的第三方”属性；其次运行 mcp_scan.py 对命令执行、文件系统写、网络外联及凭证暴露四大类风险进行扫描与打分；接着依据 LGD 三律模型交叉验证权限边界是否封闭且可追溯；最后生成包含风险评级（高/中/低）与最小授权建议的最终报告，供决策者执行准入或熔断。

证据与安全声明

• 证据：TBD（暂无评测/徽章链接；产生后在此更新实际链接）
• 版本：1.1.1
• 安全：未做独立安全扫描（TBD），结论待补；补全前不作推荐背书。