protect-mcp
Security gateway that wraps any MCP server with per-tool policies, approval gates, and optional Ed25519-signed decision receipts. Shadow mode logs every tool call without blocking; enforce mode applies block, rate-limit, and minimum-tier rules. Receipts are independently verifiable offline with no accounts needed.
README
<p align="center"> <img src="https://www.scopeblind.com/scopeblind-logo-solarin.png" width="48" /> </p>
<h1 align="center">ScopeBlind</h1>
<p align="center"> Security gateway for MCP servers.<br/> Shadow-mode logs. Per-tool policies. Optional signed receipts. </p>
<p align="center"> <a href="https://scopeblind.com">Website</a> · <a href="https://scopeblind.com/docs">Docs</a> · <a href="https://www.npmjs.com/package/protect-mcp">npm</a> · <a href="https://scopeblind.com/verify">Verify a Receipt</a> </p>
<p align="center"> <img alt="License" src="https://img.shields.io/badge/license-FSL--1.1--MIT-blue.svg" /> <img alt="npm protect-mcp" src="https://img.shields.io/npm/v/protect-mcp?label=protect-mcp&color=cb3837&logo=npm" /> <img alt="TypeScript" src="https://img.shields.io/badge/TypeScript-5.6-3178C6?logo=typescript&logoColor=white" /> </p>
protect-mcp — MCP Server Security Gateway
Wrap any stdio MCP server in one line. Start in shadow mode to see every tool call. Add a policy file to enforce per-tool rules. Generate local keys with protect-mcp init if you want signed receipts.
# Shadow mode — log every tool call, enforce nothing
npx protect-mcp -- node your-mcp-server.js
# Generate local signing keys + config template
npx protect-mcp init
# Run with policy + local signing
npx protect-mcp --policy protect-mcp.json -- node your-mcp-server.js
# Try the built-in demo
npx protect-mcp demo
What ships today
- Shadow mode (default) — logs every tool invocation with structured decision entries. Blocks nothing.
- Enforce mode — applies per-tool policies:
block,rate_limit,min_tier. - Optional local signing — when signing is configured, emits Ed25519-signed receipts alongside decision logs.
- Demo command —
npx protect-mcp demoruns a built-in 5-tool MCP server wrapped with the gateway. - Status command —
npx protect-mcp statusshows tool call stats from the local decision log. - Evidence store — file-based receipt history per agent for trust tier promotion.
- Verification — receipts verify offline with
npx @veritasacta/verifyor at scopeblind.com/verify. - No account required — local process, local config, local keys.
Current capability boundaries
- The bare
npx protect-mcp -- ...path emits logs, not signed receipts. Runprotect-mcp initfor signing. - Tier-aware policy checks are live, but manifest admission is not wired into the default CLI path. CLI sessions default to
unknownunless a host integration sets admission state programmatically. - Credential config validates env-backed references and records credential labels in logs/receipts. Generic per-call injection is adapter-specific.
- External PDP adapters (OPA, Cerbos, generic) and audit bundle helpers are exported as programmatic hooks, not fully wired into the default CLI path.
Example policy
{
"default_tier": "unknown",
"tools": {
"delete_database": { "block": true },
"write_file": { "min_tier": "signed-known", "rate_limit": "10/minute" },
"read_file": { "rate_limit": "50/minute" },
"*": { "rate_limit": "100/hour" }
},
"signing": {
"key_path": "./keys/gateway.json",
"issuer": "protect-mcp",
"enabled": true
}
}
Claude Desktop / Cursor config
{
"mcpServers": {
"my-server": {
"command": "npx",
"args": ["protect-mcp", "--policy", "protect-mcp.json", "--enforce", "--", "node", "my-server.js"]
}
}
}
Works with Claude Desktop, Cursor, VS Code — any client that speaks MCP over stdio.
The Bigger Picture
ScopeBlind produces signed, portable receipts for machine access decisions. Each receipt contains the decision, policy digest, trust tier, and timestamp — signed with Ed25519 and verifiable by anyone without calling ScopeBlind.
protect-mcp is the free, open-source entry point for MCP servers. The ScopeBlind platform adds managed signing, a real-time dashboard, and edge enforcement.
Machines need receipts. Receipts shouldn't require surveillance.
Packages
Architecture
MCP Client (Claude, Cursor, VS Code)
→ protect-mcp (stdio proxy)
→ Intercept tools/call JSON-RPC
→ Evaluate policy (allow / block / rate_limit)
→ Log decision to stderr ([PROTECT_MCP] prefix)
→ Sign receipt if signing configured ([PROTECT_MCP_RECEIPT] prefix)
→ Forward allowed calls to wrapped MCP server
License
Source-available under the Functional Source License (FSL-1.1-MIT). You may use, modify, and self-host this freely for your own projects or internal company use. You may not offer ScopeBlind (or a substantially similar service) as a hosted/managed product to third parties. After 2 years, each version automatically converts to the MIT license.
<p align="center"> Built by <a href="https://github.com/tomjwxf">Tom Farley</a> in Sydney, Australia. </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。