opzyai
Local-first security check for AI coding agents — finds hardcoded secrets, exposed .env files, git-history leaks and vulnerable dependencies (OSV), entirely on your machine. Ask your agent "is this safe to ship?" and get a Launch Readiness score with a fix for every finding.
README
Opzyai MCP — local security check for AI coding agents
Source for @opzyai/mcp — an
MCP server that scans the project in your workspace for
the mistakes that ship secrets and vulnerabilities to production, entirely on your
machine. Ask your agent "is this safe to ship?" and get a Launch Readiness score with
a fix for every finding.
Built by Opzyai — security for apps built with AI tools like Cursor, Lovable, v0 and Bolt.
Install
# Claude Code
claude mcp add opzyai -- npx -y @opzyai/mcp
// Cursor / generic MCP client
{
"mcpServers": {
"opzyai": { "command": "npx", "args": ["-y", "@opzyai/mcp"] }
}
}
What it checks
One tool — security_check({ path?, offline? }) — runs four detectors:
| Detector | Catches |
|---|---|
| Working-tree secrets | API keys/tokens hardcoded in source (OpenAI, Anthropic, Stripe, Supabase service-role, AWS, GitHub, …) |
.env exposure |
env files committed or not gitignored |
| Git-history secrets | credentials committed once and "removed" — still recoverable from history |
| Dependency CVEs | known-vulnerable packages via OSV (package-lock.json, pnpm-lock.yaml, yarn.lock) |
Detection is precision-first: an explicit allowlist keeps intentionally-public values
(Stripe pk_*, Supabase anon keys) from ever being flagged.
Privacy
Everything runs locally over stdio. The only network call is the OSV dependency check —
package names + versions only, never your code — and offline: true disables even
that.
Repository layout
This is the public source mirror of the local scanner; it is developed inside the private Opzyai monorepo and synced here on each release, byte-identical.
packages/
mcp-local/ @opzyai/mcp — the MCP server published to npm
detectors/ @appsec/detectors — shared secret-detection patterns + allowlist
core/ @appsec/core — trimmed shim (shared types only; the full package is server-side)
Develop
pnpm install
pnpm typecheck && pnpm test # vitest, all packages
pnpm build # tsup → packages/mcp-local/dist/cli.js
Requires Node >= 20 and git on PATH (for the git-history detector's tests).
Related
- Free URL scan (no account): paste your deployed URL at
opzyai.com/scan — passive check for
leaked client-bundle keys, exposed
.env/.git/source maps, missing headers. - Hosted Pro MCP: deep scans of repos you own (dependency CVEs, SAST, git-history
secrets) plus
propose_fix— the exact change for your agent to apply: opzyai.com/mcp.
License
MIT © Opzyai
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。