package-verify-mcp
Verifies npm package names for safety, detecting typosquats and nonexistent packages before your coding agent installs them.
README
package-verify-mcp
Catch hallucinated and typosquatted npm packages before your coding agent installs them.
An MCP server that verifies an npm package name is real and safe before Claude, Cursor, or any agent runs npm install. AI agents routinely invent plausible-sounding package names ("slopsquatting"), and attackers pre-register those exact names with malware — one hallucinated package (react-codeshift) spread into 237 repos. This server is the guardrail.
Why this exists
An agent confidently writes import x from "some-plausible-pkg" and then installs it. If that name doesn't exist, best case you get an error; worst case a squatter is waiting there with a postinstall script. Checking is one registry call — this makes the agent do it every time.
Tools
| Tool | What it does |
|---|---|
verify_package |
Verdict for one name — SAFE / CAUTION / SUSPICIOUS / DOES_NOT_EXIST — from existence, first-publish age, weekly downloads, deprecation, and edit-distance to popular packages (typosquat detection). |
verify_packages |
Batch check — pass every dependency in a proposed change; flagged names sort to the top. |
Uses only the public npm registry + downloads API. No key.
What the verdicts mean
- ✅ SAFE — exists, established (age + downloads), not a near-miss of a popular name.
- ⚠️ CAUTION — exists but young, low-traffic, deprecated, or a similar-but-established name. Confirm intent.
- 🛑 SUSPICIOUS — one or two edits from a popular package and young/low-traffic: a classic typosquat. Don't install without confirming.
- ❌ DOES_NOT_EXIST — not on npm. Hallucination signature — do not install; suggestions for the likely intended name are included.
Quick start
npx package-verify-mcp
Claude Code
claude mcp add package-verify -- npx -y package-verify-mcp
Claude Desktop / Cursor / Windsurf / any MCP client
{
"mcpServers": {
"package-verify": {
"command": "npx",
"args": ["-y", "package-verify-mcp"]
}
}
}
Pairs well with a rule like "Before adding any npm dependency, verify it with package-verify."
Example prompts
- "Verify
react-hook-formsbefore you install it." (→ 🛑 typosquat ofreact-hook-form) - "Check all the packages in this package.json diff with package-verify."
- "Is
@types/node-fetcha real package?"
Config
| Env var | Default | Purpose |
|---|---|---|
NPM_REGISTRY |
https://registry.npmjs.org |
Registry for existence/metadata. |
NPM_API |
https://api.npmjs.org |
Downloads API for weekly-download signal. |
How it works
package name
│
├─ registry GET ── exists? first-publish date, latest, deprecated, repo
├─ downloads API ── weekly downloads (traffic/trust signal)
└─ edit-distance vs bundled popular-package list ── typosquat detection
└─► SAFE / CAUTION / SUSPICIOUS / DOES_NOT_EXIST + reasons + suggestions
Caveats
- This is a hallucination / typosquat tripwire, not a full security scanner — it won't analyze package contents (use Socket/npq for deep supply-chain analysis). It answers "is this name real and plausibly the one you meant."
- The popular-package reference list is curated, not exhaustive; extend
src/popular.tsfor your stack.
Develop
npm install
npm run build
node dist/index.js
License
MIT © Anicodeth
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。