HaizhuOpsPanel

A visual server operations and AI application deployment platform for independent developers, technical teams, and enterprises. The external landing page is built with Next.js, React, and Tailwind CSS; the backend provides multi-host asset, site certificate, database, container, task, backup, security audit, and application marketplace capabilities over SSH.

v2.1.0 adds MCP processing capabilities, v2.2.0 extends GitHub deployment: The panel's operations capabilities are packaged as 16 standard MCP tools (service inspection, container management, disk diagnostics, application deployment, GitHub open-source project search and one-click deployment, etc.), supporting two usage modes — direct connection from external MCP clients (Claude Desktop / Cursor / VS Code, etc.), or natural language conversation via the panel's built-in AI operations assistant. Application deployment is no longer limited to the 80 apps in the built-in store; you can search the latest GitHub projects and pull them directly for deployment. See docs/mcp.md for details.

Quick Start (Operations Panel + MCP / AI Capabilities)

npm install
npm run start:legacy        # 启动运维面板 http://localhost:8899（默认 admin / admin123）
npm run start:mcp           # （可选）启动 MCP 服务器（stdio，供外部 AI 客户端接入）

To enable the panel's built-in AI assistant: create .env in the project root and write AI_API_KEY=sk-xxx (optionally AI_BASE_URL, AI_MODEL), then restart the panel and describe your operations needs in natural language in the "AI Operations Assistant" on the left. For external MCP client integration configuration, see docs/mcp.md and scripts/mcp-config.json.

Related MCP server: Dokploy MCP Server

Tech Stack

• Next.js 15 (App Router)

• React 19

• Tailwind CSS 4

• Lucide React

• The original Express / SSH / WebSocket operations backend remains in server/

Local Development

npm install
npm run dev

Visit http://localhost:3000.

Production Build

npm run build
npm start

Quality Checks

npm run lint
npm run build

Project Structure

├─ app/                     # Next.js 页面、元信息与全局设计样式
├─ components/              # 落地页组件与交互式产品演示
├─ design-system/haizhuopspanel/ # ui-ux-pro-max 生成的企业设计系统
├─ legacy-public/           # 升级前的原生面板前端，完整保留（含 AI 运维助手）
├─ server/                  # Express API、SSH 与 WebSocket 后端
│  ├─ mcp-tools.js          # MCP 工具层（16 个运维工具，共用执行入口）
│  ├─ mcp-server.js         # stdio MCP 服务器（外部 AI 客户端接入）
│  └─ ai.js                 # 面板内置 AI 助手（OpenAI 兼容 function calling）
├─ scripts/mcp-config.json  # 各 MCP 客户端接入配置示例
├─ docs/mcp.md              # MCP 与 AI 助手使用指南
├─ postcss.config.mjs       # Tailwind CSS PostCSS 配置
└─ package.json

MCP Processing Capabilities (v2.1.0 / v2.2.0)

Users describe their needs in natural language, and the AI executes CLI commands on remote hosts through MCP tools to complete service inspection and operations deployment:

• 16 standard MCP tools: host list, system overview, systemd service inspection/start-stop, Docker container list/operations/logs, disk diagnostics, process ranking, app store search, app deployment/uninstall, GitHub project search/deployment, audit queries

• GitHub arbitrary repository deployment (v2.2.0): apps not available in the built-in store automatically fall back to GitHub search (sorted by Stars); after confirmation, one-click deployment — automatically installs Git/Docker, intelligently detects docker-compose / Dockerfile, monorepo supports subdirectories, and re-running updates to the latest version

• Two entry points:

  • External MCP clients (Claude Desktop, Cursor, Cline, VS Code, WorkBuddy, etc.) connect via npm run start:mcp

  • The panel's built-in "AI Operations Assistant" chat interface (any compatible API such as OpenAI / DeepSeek / Zhipu, etc.)

• Security design: host credentials are AES-256-GCM encrypted and never leave the database, all tool calls are written to audit logs, run_command is disabled by default, and the AI confirms before executing any change operations

Configuration methods and examples are in docs/mcp.md.

Design and Experience

• Professional, clean dark-green operations visual style, using semantic status colors rather than purely decorative colors

• Responsive adaptation for 375px, 768px, 1024px, and large desktop screens

• Mobile navigation and switchable product workflow demos

• Touch targets of 44px or larger, clear keyboard focus, sensible heading hierarchy

• Supports prefers-reduced-motion, avoiding unnecessary dynamic effects

• Statically pre-rendered homepage to keep client load small

Enterprise Operations Capabilities

• Multi-host asset fingerprinting and abnormal service inspection

• Nginx, Apache, Caddy site and SSL certificate expiration inventory

• MySQL, PostgreSQL, Redis, MongoDB instance discovery

• Docker container, image, network, and Compose application management

• Unified task overview for Cron, system Cron, and systemd Timers

• Backup freshness inspection and pre-change configuration snapshots

• SSH, account, sudo, firewall, port, and failed login security baseline

• SSH public key fingerprint and file permission audit

• JSONL audit records for logins, host changes, application deployments, and operations actions

New AI Application Templates

The app marketplace adds DeerFlow 2.0, Wegent, InsForge, OpenAgent, RuFlo, and CLASP, while retaining the original Open WebUI, Ollama, RAGFlow, LobeChat, New API, and other templates. All new projects use the official public Docker, Compose, or installation script paths from the projects themselves.

Research basis is in Product and Feature Research.

Original Operations Panel

The pre-upgrade frontend has been moved to legacy-public/, and the backend implementation remains in server/. To continue running the old panel, run:

npm run start:legacy

The old service has been changed to load static assets from legacy-public/, so it can be maintained in parallel with the new Next.js landing page.

License

MIT