DepShield MCP
Acts as a security checkpoint for AI coding agents by intercepting package installations to verify existence, check against CVE databases, and block vulnerable or hallucinated dependencies before they reach your codebase. Provides seven security tools including pre-install gates, full project audits, safe version recommendations, and deep transitive dependency scanning for npm and PyPI packages.
README
<p align="center"> <img src="https://img.shields.io/badge/🛡️_DepShield-MCP_Server-blue?style=for-the-badge" alt="DepShield MCP" /> </p>
<h3 align="center">Real-time dependency security for AI coding agents</h3>
<p align="center"> <a href="https://www.npmjs.com/package/depshield-mcp"><img src="https://img.shields.io/npm/v/depshield-mcp?style=flat-square&color=cb3837&label=npm" alt="npm version" /></a> <a href="https://www.npmjs.com/package/depshield-mcp"><img src="https://img.shields.io/npm/dm/depshield-mcp?style=flat-square&color=blue&label=downloads" alt="npm downloads" /></a> <a href="https://github.com/devanshkaria88/depshield-mcp/blob/main/LICENSE"><img src="https://img.shields.io/github/license/devanshkaria88/depshield-mcp?style=flat-square" alt="license" /></a> <a href="https://github.com/devanshkaria88/depshield-mcp"><img src="https://img.shields.io/github/stars/devanshkaria88/depshield-mcp?style=flat-square" alt="GitHub stars" /></a> <img src="https://img.shields.io/badge/node-%3E%3D22-brightgreen?style=flat-square" alt="node version" /> </p>
<p align="center"> DepShield intercepts every package your agent tries to install, verifies it exists, checks it against CVE databases, and blocks vulnerable or hallucinated dependencies — <strong>before they touch your codebase</strong>. </p>
<p align="center"> Works with <strong>Cursor</strong> · <strong>Claude Code</strong> · <strong>Windsurf</strong> · <strong>Antigravity</strong> · <strong>Cline</strong> · any MCP-compatible IDE </p>
🤔 Why
AI coding agents pull packages from stale training data. They install outdated versions with known CVEs. They hallucinate package names that don't exist, opening the door to typosquatting attacks. They default to deprecated libraries when better alternatives exist.
There's no checkpoint between "agent decides to use a package" and "package lands in your project." DepShield is that checkpoint.
⚡ What It Does
Seven security tools exposed over the Model Context Protocol:
| Tool | What it does |
|---|---|
check_dependency |
Pre-install gate — verifies a package exists on the registry and has no known CVEs. Your agent calls this before every install. |
audit_project |
Scans an entire package.json or requirements.txt and returns a full vulnerability audit report. |
find_safe_version |
Finds the newest version of a package with zero known vulnerabilities. |
get_advisory_detail |
Deep dive into a specific CVE/GHSA — full description, affected versions, fix info, references. |
check_npm_health |
Package health report card: weekly downloads, last publish date, maintainer count, license, deprecation status. Scored 0–100. |
suggest_alternative |
Finds better packages when one is vulnerable, deprecated, or abandoned. |
deep_scan |
Scans a package's transitive dependency tree for vulnerabilities, typosquats, and suspicious patterns. |
Plus a depshield://status resource and a security_review prompt template for guided full-project audits.
Zero API keys required. DepShield uses free, open APIs: npm Registry, OSV.dev (Google's open vulnerability database), and PyPI.
🚀 Quick Start
Option 1: npx (no install)
Add to your IDE's MCP configuration and you're done. No global install needed.
Option 2: Clone and build
git clone https://github.com/devanshkaria88/depshield-mcp.git
cd depshield-mcp
npm install
npm run build
🔌 IDE Setup
<details> <summary><strong>Cursor</strong></summary>
Add to ~/.cursor/mcp.json (global) or .cursor/mcp.json (project-level):
{
"mcpServers": {
"depshield": {
"command": "npx",
"args": ["-y", "depshield-mcp"]
}
}
}
Or if you cloned the repo locally:
{
"mcpServers": {
"depshield": {
"command": "node",
"args": ["/absolute/path/to/depshield-mcp/dist/index.js"]
}
}
}
Optionally, copy .cursor/rules/dep-shield.mdc into your project's .cursor/rules/ directory. This rule forces the Cursor agent to call check_dependency before every package install.
</details>
<details> <summary><strong>Claude Code</strong></summary>
claude mcp add depshield -- npx depshield-mcp
Or from a cloned repo:
claude mcp add depshield -- node /absolute/path/to/depshield-mcp/dist/index.js
</details>
<details> <summary><strong>Windsurf</strong></summary>
Add to your Windsurf MCP configuration:
{
"mcpServers": {
"depshield": {
"command": "npx",
"args": ["-y", "depshield-mcp"]
}
}
}
</details>
<details> <summary><strong>Any MCP-compatible tool</strong></summary>
DepShield uses stdio transport. Any tool that supports MCP over stdio can use it:
npx depshield-mcp
</details>
💡 Usage Examples
Once connected, your AI agent has access to all seven tools. Try these prompts:
<table> <tr> <td width="50%">
Pre-install check (automatic with .mdc rule)
"Add lodash for deep cloning"
Agent calls check_dependency → finds CVE → auto-upgrades to safe version
</td> <td width="50%">
Block hallucinated packages
"Install react-super-utils-pro for state management"
Agent calls check_dependency → package doesn't exist → blocks install, suggests alternatives
</td> </tr> <tr> <td>
Full project audit
"Run a security audit on this project's dependencies"
Agent calls audit_project on package.json → returns full vulnerability report
</td> <td>
Package health check
"Is this package well maintained?"
Agent calls check_npm_health → returns health score (0–100) with breakdown
</td> </tr> <tr> <td>
Supply chain deep scan
"Deep scan express for transitive vulnerabilities"
Agent calls deep_scan → scans dependency tree, flags suspicious patterns
</td> <td>
Advisory deep dive
"Tell me more about GHSA-jf85-cpcp-j695"
Agent calls get_advisory_detail → returns full CVE details, remediation info
</td> </tr> </table>
📖 Tools Reference
<details> <summary><code>check_dependency</code> — Pre-install gate</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
name |
string | yes | — | Package name (e.g., lodash, express) |
version |
string | no | latest | Specific version to check |
ecosystem |
npm | pypi |
no | npm |
Package ecosystem |
Returns: ✅ SAFE, ⚠️ VULNERABLE (with fix version), 🚫 BLOCKED (doesn't exist), or ⚠️ CANNOT VERIFY (registry unreachable).
</details>
<details> <summary><code>audit_project</code> — Full manifest audit</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
filePath |
string | yes | — | Path to package.json or requirements.txt |
includeDevDependencies |
boolean | no | true |
Include devDependencies in scan |
Returns: Full audit report with summary stats, per-dependency vulnerability breakdown, severity counts, and risk verdict.
</details>
<details> <summary><code>find_safe_version</code> — Safe version finder</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
name |
string | yes | — | Package name |
ecosystem |
npm | pypi |
no | npm |
Package ecosystem |
Returns: The newest stable version with zero known vulnerabilities, with comparison of all checked versions.
</details>
<details> <summary><code>get_advisory_detail</code> — CVE/GHSA deep dive</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
vulnId |
string | yes | — | Vulnerability ID (e.g., GHSA-jf85-cpcp-j695, CVE-2021-23337) |
Returns: Full advisory with summary, severity, CVSS score, affected versions, fix versions, and reference links.
</details>
<details> <summary><code>check_npm_health</code> — Package health scoring</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
name |
string | yes | — | npm package name |
Returns: Health report card (0–100 score) based on: publish recency, weekly downloads, license, repository presence, deprecation status, and maintainer count.
</details>
<details> <summary><code>suggest_alternative</code> — Alternative package finder</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
name |
string | yes | — | Package to find alternatives for |
reason |
string | no | — | Why an alternative is needed |
Returns: Top 3 alternative packages with downloads, npm score, and publish date.
</details>
<details> <summary><code>deep_scan</code> — Transitive dependency scanner</summary>
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
name |
string | yes | — | Package to deep scan |
version |
string | no | latest | Specific version to scan |
depth |
1 | 2 |
no | 1 |
1 = direct deps, 2 = deps of deps |
Returns: Dependency tree with vulnerability flags, suspicious pattern detection (newly added deps, nonexistent packages, low-download typosquat candidates), and risk verdict.
</details>
🔒 Cursor Rule (Optional)
The .cursor/rules/dep-shield.mdc file included in this repo forces the Cursor agent to automatically call check_dependency before every package install. Copy it to your project:
mkdir -p .cursor/rules
cp node_modules/depshield-mcp/.cursor/rules/dep-shield.mdc .cursor/rules/
This turns DepShield from a tool the agent can use into a gate the agent must pass through.
🧪 Testing
MCP Inspector (interactive UI for testing tools):
npm run inspect
Or:
npx @modelcontextprotocol/inspector node dist/index.js
Raw stdio (quick verification):
printf '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}\n{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"check_dependency","arguments":{"name":"lodash","version":"4.17.20"}}}\n' | node dist/index.js 2>/dev/null
🌐 APIs Used
All free, no API keys required:
| API | What it provides |
|---|---|
| npm Registry | Package existence, versions, metadata, search |
| npm Downloads API | Weekly download counts |
| OSV.dev | Open source vulnerability database (Google) |
| PyPI | Python package metadata |
📁 Project Structure
depshield-mcp/
├── src/
│ ├── index.ts # MCP server entry — tool/resource/prompt registration
│ ├── cache.ts # In-memory TTL cache (5 min)
│ ├── utils.ts # Severity parsing, version sorting, fetch helpers
│ ├── apis/
│ │ ├── npm-registry.ts # npm registry, search, downloads
│ │ ├── osv.ts # OSV.dev vulnerability queries
│ │ └── pypi-registry.ts # PyPI JSON API
│ └── tools/
│ ├── check-dependency.ts # Pre-install gate
│ ├── audit-project.ts # Full manifest audit
│ ├── find-safe-version.ts # Safe version finder
│ ├── get-advisory-detail.ts # CVE/GHSA deep dive
│ ├── check-npm-health.ts # Package health scoring
│ ├── suggest-alternative.ts # Alternative package finder
│ └── deep-scan.ts # Transitive dependency scanner
├── .cursor/rules/
│ └── dep-shield.mdc # Cursor agent rule (alwaysApply)
├── package.json
├── tsconfig.json
├── LICENSE
└── README.md
📋 Requirements
- Node.js 22 or later
- An MCP-compatible AI coding tool
🤝 Contributing
Contributions are welcome. Please open an issue first to discuss what you'd like to change.
Areas where help would be appreciated:
- Additional ecosystem support (Cargo, Go modules, Maven)
- Improved CVSS vector string parsing
- Integration tests with mocked API responses
- GitHub Actions CI pipeline
📄 License
MIT — see LICENSE for details.
<p align="center"> Built by <a href="https://github.com/devanshkaria88">Devansh Karia</a> </p>
<p align="center"> <a href="https://buymeacoffee.com/devanshkaria88"> <img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" alt="Buy Me A Coffee" height="50" width="217" /> </a> </p>
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。