DynamicAgent2UI 🖥️🤖

An OS-Native Floating Dialog & Form UI Canvas that exposes interactive desktop widgets to AI agents using the Model Context Protocol (MCP).

With DynamicAgent2UI, any MCP-compatible AI agent (such as Gemini Antigravity, Claude Desktop, or VS Code Cline) can render high-fidelity, native-looking dialogs and settings forms directly on the user's desktop screen, block execution, and retrieve the user's interaction (button clicks, form inputs) as the tool response.

✨ Features

• 🖥️ OS-Native Styles: Renders native-looking dialogs and forms matching macOS, Windows 11 (Fluent), and Android (Material 3).

• 📐 Dynamic Window Resizing: Utilizes ResizeObserver and Electron IPC to resize the frameless window to the exact dimensions of the dialog content (+ drop shadow padding), allowing clicks on transparent areas to pass through to background apps.

• 🕳️ Opacity Correction: Uses solid, opaque fills (bg-white & bg-[#1e1e1e]) preventing desktop wallpapers from bleeding through and keeping text/controls highly legible.

• 🎛️ Draggable Regions: Allows the user to click and drag the dialog background to position it anywhere on their screen, while keeping buttons and inputs interactive.

• 🔌 Unified Self-Starting MCP: The MCP server automatically checks port 3000 and launches the Next.js backend and Electron app in the background when the agent connects. Zero manual CLI commands required!

Related MCP server: mcp-interactive-ui-server

🛠️ Tech Stack

• Frontend: Next.js (App Router, Tailwind CSS, TypeScript, Zod)

• Desktop Wrapper: Electron (Frameless, Transparent, IPC, Node Integration)

• Protocol: Model Context Protocol (MCP JSON-RPC 2.0 over stdio)

🚀 Getting Started

1. Prerequisites

• Node.js (v18 or higher recommended)

• npm (or pnpm/yarn)

2. Installation & Environment Configuration

Clone the repository, enter the directory, and install dependencies:

git clone https://github.com/your-username/DynamicAgent2UI.git
cd DynamicAgent2UI
npm install

Create a .env file in the root directory to configure the AI agent's chat interface (optional):

# Optional: Setup a custom LLM endpoint (OpenAI compatible) for the built-in control panel chat
CUSTOM_LLM_BASE_URL=https://api.your-provider.com/v1
CUSTOM_LLM_API_KEY=your-api-key
CUSTOM_LLM_MODEL=your-model-name

🔌 Using with an MCP Client (e.g. Gemini, Claude Desktop, Cline)

To integrate DynamicAgent2UI with your agent, add it to your client's MCP configuration settings file (e.g., claude_desktop_config.json or cline_mcp_settings.json):

{
  "mcpServers": {
    "DynamicAgent2UI": {
      "command": "node",
      "args": ["C:/Workspace/OpenUI/mcp-server.js"]
    }
  }
}

Note: Replace the absolute path in args with your cloned repository path.

Once configured and restarted, the agent will have access to the following tools:

Tool: show_dialog

Displays a native OS dialog and blocks until a button is clicked.

• primaryButton (required): Text label (e.g. "OK", "Save").

• secondaryButton / cancelButton (optional): Alternative button labels.

• title / message (optional): Title and body description.

• icon (optional): "info" | "warning" | "error" | "question" | "success".

• platform (optional): "macos" | "windows" | "android". Defaults to auto-detecting the host OS.

• theme (optional): "light" | "dark". Default is "light".

• inputPlaceholder (optional): Adds a text input box.

Tool: show_form

Displays a native multi-input form panel.

• title (required): Form header title.

• fields (required): Comma-separated labels and types, e.g. "Username: text, Age: number, Role: select(Admin|User), Active: checkbox".

• submitButton (required): Button label.

• platform / theme (optional): OS style and color theme.

🏃 Running Manually (Optional)

If you wish to run the web interface or desktop window manually without the MCP server:

• Run the Next.js development server:

  npm run dev

• Run the Electron desktop window:

  npm run desktop

• Build for production:

  npm run build