autogui-mcp

[中文](https://github.com/stellariums/AutoGUI/blob/master/README_CN.md) | English # AutoGUI AI-driven screen automation MCP Server. Send natural language tasks, and the internal AI captures screenshots, analyzes them, and performs mouse/keyboard actions autonomously. ## Architecture ``` MCP Client (Claude Code, Claude Desktop, Cursor, etc.) | stdio v server.py (FastMCP async orchestration loop) | v agent.py (ScreenAgent toolkit: capture, execute, parse, safety) ``` ## Quick Start ```bash git clone https://github.com/stellariums/AutoGUI.git cd AutoGUI pip install -r requirements.txt ``` ## Configuration AutoGUI supports layered configuration: **environment variables** (highest priority) > **config.json** > **defaults**. ### Option A: Environment Variables Only (Simplest) Only 3 variables needed to get started: ```bash set AUTOGUI_API_KEY=your-api-key set AUTOGUI_BASE_URL=https://api.openai.com/v1 set AUTOGUI_MODEL=gpt-4o ``` Or copy `.env.example` to `.env` and pass via your MCP client config (see below). ### Option B: Config File ```bash cp config.json.example config.json ``` Edit `config.json` — only the `api` section is required, everything else has sensible defaults: ```json { "api": { "base_url": "https://your-api-endpoint/v1", "api_key": "your-api-key", "model": "your-model-name" } } ``` <details> <summary>Advanced config options</summary> ```json { "screen": { "max_width": 1280, "max_height": 720, "allowed_region": null }, "agent": { "max_iterations": 20, "delay_between_actions": 1.0, "max_history_rounds": 10 }, "safety": { "enable_confirmation": true, "fallback_action": "block", "dangerous_keys": ["delete", "backspace", "escape"], "dangerous_hotkeys": [["ctrl", "w"], ["alt", "f4"]], "dangerous_patterns": ["rm ", "del ", "format ", "shutdown"] } } ``` </details> ## MCP…