mcp-trust-demo
An intentionally vulnerable MCP server designed as a live demo target for the MCP Trust security scanner. It contains deliberate insecure patterns to demonstrate scanning capabilities.
README
Vulnerable MCP Server — a live demo for MCP Trust
⚠️ This repository is intentionally vulnerable
It contains a deliberately insecure MCP server that exists only as a live demo target for the MCP Trust scanner. Do not install, run, import, or copy this code. There is no real functionality here — every risky pattern is on purpose. The exfiltration URL (
attacker.example) is a reserved, non-routable domain; there are no real secrets.
What this repo demonstrates
That wiring MCP Trust into a repo's CI takes one line, and that it catches real issues in an MCP server before an agent ever connects to it:
# .github/workflows/mcp-trust.yml
- uses: actions/checkout@v4
- uses: SteveMonsway/mcp-trust@v1
with:
fail-on: high
upload-sarif: true # → GitHub Security / Code Scanning alerts
comment-pr: true # → summary comment on pull requests
That's the entire integration. Full inputs/outputs: the MCP Trust GitHub Action.
What MCP Trust flags here
Scanning server.js + package.json produces
Decision: BLOCK on the following evidence:
| Rule | Severity | What |
|---|---|---|
MCP-CODE-001 |
critical | child_process.exec with caller-controlled input |
MCP-META-001 |
high | Tool description with an instruction-override phrase ("ignore previous instructions") |
MCP-META-002 |
high | Tool description with a concealment phrase ("do not tell the user") |
MCP-CODE-005 |
medium | Arbitrary filesystem read at a caller-supplied path |
MCP-SUPPLY-003 |
medium | postinstall script runs arbitrary code on npm install |
MCP-CODE-007 |
low | Secret-like environment variable access (GITHUB_TOKEN) |
MCP-SUPPLY-001 |
low | No linked source repository |
See it live
- Actions → latest MCP Trust run
- Security → Code Scanning → the SARIF alerts uploaded by the scan
- Pull Requests → the open demo PR shows the MCP Trust comment and a red ✗ check — that's the scanner gating a change
Why is the check red on PRs but green on main?
The target is vulnerable, so the scan is always BLOCK. On a pull request the job
fails on purpose (fail-on: high) — that's MCP Trust blocking the merge, exactly what
you want in review. On push to main the step is set to continue-on-error, so the
branch badge stays green while the SARIF alerts still populate the Security tab. A green
run here would mean the scanner missed the vulnerabilities.
Try it against your own MCP server
Point the scanner at any public repo, npm package, or local path:
npx @mcp-trust/cli scan github:owner/repo
Or add the Action (the one-liner above) to your own repository's
.github/workflows/. More: MCP Trust
· public benchmark of 452 real MCP servers.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。