benni-operator-gateway
An open-source MCP gateway for AI operators, connecting any LLM to a hot-reloadable connector registry with production-grade approval gates and a built-in control plane for session management.
README
Benni Operator Gateway
Open Source MCP Gateway for AI Operators — connect any LLM to your autonomous agent stack with production-grade approval gates, hot-reload connector registry, and a built-in Control Plane.
What is this?
Benni Operator Gateway is the open source core of the Benni OS autonomous AI operator stack. It implements the Model Context Protocol (MCP) as a production-grade HTTP gateway — letting any MCP-compatible AI client (Claude Desktop, Cursor, custom agents) connect to a unified, extensible set of tools and connectors.
Built for AI builders who want:
- A self-hosted alternative to managed tool-calling APIs
- Full control over what tools their agents can access
- Human-in-the-loop approval gates before sensitive actions execute
- A connector registry that hot-reloads without downtime
Architecture
graph TD
A[AI Client\nClaude / Cursor / Agent] -->|MCP over HTTP| B[Benni Operator Gateway]
B --> C[Connector Registry]
C --> D[JARVAS-2 Connector]
C --> E[GitHub Connector]
C --> F[Custom Connectors]
B --> G[Approval Gate Middleware]
G -->|webhook| H[Human Operator]
B --> I[Control Plane]
I --> J[Session State]
I --> K[Decision Ledger]
I --> L[Atomic Checkpoints]
Quick Start
# 1. Clone
git clone https://github.com/benni-os/benni-operator-gateway
cd benni-operator-gateway
# 2. Install
npm install
# 3. Configure
cp .env.example .env
# Edit .env with your API keys
# 4. Run
npm run dev
Your MCP gateway is now live at http://localhost:3000.
Built-in Connectors
| Connector | Status | Description |
|---|---|---|
jarvas2 |
✅ Stable | JARVAS-2 autonomous agent — run tasks, queue jobs, set objectives |
github |
✅ Stable | Full GitHub API via MCP — issues, PRs, code, reviews |
control-plane |
🔧 Beta | Session state, atomic checkpoints, decision ledger |
Project Structure
src/
├── app.ts # Server bootstrap
├── index.ts # Entrypoint
├── config/ # Environment & provider config
├── connectors/ # MCP connector modules
│ ├── jarvas2/ # JARVAS-2 connector
│ └── github/ # GitHub MCP connector
├── lib/ # Core utilities
├── middleware/ # Auth, approval gate, rate limiting
├── registry/ # Hot-reload connector registry
├── routes/ # HTTP endpoints
├── services/ # LLM dispatch, session management
└── types/ # TypeScript contracts
Approval Gate
The gateway includes a built-in human approval middleware that intercepts sensitive tool calls before execution:
// Any tool marked requires_approval: true will pause and
// send a webhook to your configured APPROVAL_WEBHOOK_URL
// before proceeding. The agent waits for your response.
{
"action": "delete_production_database",
"reason": "Cleanup task requested by agent",
"payload": { ... },
"approve_url": "https://your-gateway/approve/req_abc123",
"reject_url": "https://your-gateway/reject/req_abc123"
}
Roadmap
- [x] MCP HTTP gateway core
- [x] JARVAS-2 connector
- [x] GitHub connector
- [x] Human approval gate
- [ ] npm package
@benni-os/operator-gateway - [ ] Control Plane SDK
- [ ] Docker image + one-click Railway/Render deploy
- [ ] Web dashboard (runs + decisions + health)
- [ ] Benni OS Cloud (hosted)
Contributing
See CONTRIBUTING.md. All connector contributions welcome.
Part of Benni OS
This gateway is the open source foundation of Benni OS — a full autonomous AI operator stack. Follow the project:
- GitHub Org: github.com/benni-os
- Built by Benni Alencar
MIT License — use freely, contribute back.
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。