[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_123":124},"routes/modules.$creator.$moduleId",{"_14":15,"_16":17,"_18":19,"_20":21},"routes/modules.$creator.$moduleId.endpoints",{"_11":12,"_13":-5},"error","Module not found","tools","module",{"_20":22,"_23":24,"_25":26,"_27":28,"_29":30,"_31":32,"_33":34,"_35":26,"_36":37,"_38":-5,"_39":-5,"_40":-5,"_41":42,"_43":44,"_45":46,"_47":48,"_49":50,"_51":52,"_53":54,"_55":56,"_57":-5,"_58":50,"_59":60,"_61":62,"_63":-5,"_64":-5,"_65":-5,"_66":-5,"_67":68,"_69":70,"_71":72,"_73":74,"_75":76,"_77":-5,"_78":-5,"_79":-5,"_80":81,"_82":-5,"_83":44,"_84":85,"_86":87,"_88":89,"_90":91,"_92":93,"_94":95},"apiUrl","https://odel.app/api","creator","xeakaes","moduleId","computer-use-for-all-agents","@xeakaes/computer-use-for-all-agents","displayName","Screen Control","tagline","Safety-gated computer use for AI agents: screen OCR and vision, mouse, keyboard, window control.","iconKey","https://avatars.githubusercontent.com/u/194935315?v=4","category","local","status","published","visibility","public","metaDescription","deploymentType","external","workerName","bindingName","mcpConfig","pricingModel","free","priceCredits",0,"creatorUserId","system:external-import","createdAt",1789714040000,"updatedAt",1790464319000,"sourceRegistry","official-mcp-registry","sourceUrl","https://registry.modelcontextprotocol.io/v0/servers?search=io.github.Xeakaes%2Fcomputer-use-for-all-agents","sourceVerified",true,"sourceMissingSince","importedAt","upstreamStars",2,"upstreamUpdatedAt",1789683517000,"detectedTransport","detectedAuthType","probeStatus","lastProbedAt","githubRepoUrl","https://github.com/Xeakaes/computer-use-for-all-agents","githubRepoFullName","Xeakaes/computer-use-for-all-agents","releaseId","ext-xeakaes-computer-use-for-all-agents","version","1.0.0","descriptionMarkdown","# 🖥️ Screen Control\n\n

\n $\"screen-control$ \n

\n $\"License:$ \n $\"Platform\"$ \n $\"Python\"$ \n $\"CI:$ \n $\"MCP\"$ \n $\"Listed$ \n

\n\n> A local remote-control system for AI agents: watch your computer's screen\n> **live** and send **mouse/keyboard commands** to it. Everything runs on your\n> own machine — no data ever leaves it, no cloud middleman.\n\n

\n $\"An$ \n

\n\n

\n _{Every keystroke and every screenshot in this GIF went through the API —\n the agent never touched a physical keyboard.}\n

\n\n---\n\n## Why Screen Control?\n\nAI agents today can write code and call APIs — but they can't *see* or *touch*\nyour desktop. Screen Control gives any agent general-purpose computer use over\na clean, safety-gated HTTP/MCP interface:\n\n- **Perceive** — OCR for text, single frames or a live MJPEG stream for\n vision-capable models, and a text-only diff endpoint for models that can't\n consume images at all.\n- **Act** — absolute and relative mouse, Unicode-safe keyboard, window\n management, background (focus-free) control, virtual desktops.\n- **Stay safe** — token auth, blocked deadly shortcuts, focus guard, a\n stuck-input watchdog and an emergency failsafe are all enforced server-side,\n no matter how confused the agent gets.\n\nOne process, zero configuration, works with any language that can speak HTTP —\nor natively through MCP in Claude Desktop, Cursor, VS Code and cloud agents.\n\n### Performance Is Agent-Bound\n\nScreen Control is the **perception and actuation layer** — the eyes and hands.\nThe effective speed and capability of any agent using it are bounded by that\nagent itself and by the environment it runs in:\n\n- **Thinking speed** — one action per agent \"turn\": the perceive → plan →\n act → verify loop lives in the agent, so model inference latency and\n reasoning depth directly set the pace. The API itself adds only\n milliseconds per call.\n- **Context capacity** — screen readings (OCR text, frames, diffs) consume\n the agent's context window; a larger window means more situational\n awareness before verification degrades.\n- **Runtime environment** — network latency, MCP/HTTP round-trip overhead,\n tool-call limits and hosting constraints all stack on top of the loop.\n\nIn practice this means: the same repo makes a fast reasoning model fast and\ncapable, and makes a slow model slow — the toolchain is not the bottleneck.\nReal-time or action-heavy tasks need an agent with fast inference and tight\ntool-loop latency; slower agents should prefer deliberate, verification-heavy\ntasks.\n\n---\n\n## Table of Contents\n\n- [Why Screen Control?](#why-screen-control)\n- [Features](#features)\n- [Architecture](#architecture)\n- [Installation](#installation)\n- [Quick Start](#quick-start)\n- [API Reference](#api-reference)\n - [Authentication](#authentication)\n - [Screen Capture](#screen-capture)\n - [Mouse Control](#mouse-control)\n - [Keyboard Control](#keyboard-control)\n - [OCR (Screen Reading)](#ocr-screen-reading)\n - [Vision Access (Image Models)](#vision-access-image-models)\n - [Window Management](#window-management)\n - [Focus-Free (Background) Control](#focus-free-background-control)\n - [Virtual Desktops](#virtual-desktops)\n - [Game Mode](#game-mode)\n - [Safety Endpoints](#safety-endpoints)\n- [🤖 For AI Agents](#-for-ai-agents)\n- [MCP Support (One-Click Cloud Agents)](#mcp-support-one-click-cloud-agents)\n- [Security Model](#security-model)\n- [Game Mode Guide](#game-mode-guide)\n- [Vision Access Guide](#vision-access-guide)\n- [Troubleshooting](#troubleshooting)\n- [Testing](#testing)\n- [Contributing](#contributing)\n- [License](#license)\n\n---\n\n## Features\n\n| Feature | Description |\n|---|---|\n| 🖼️ **Live screen feed** | Continuously refreshing screenshot in the browser |\n| 🖱️ **Mouse control** | Click, right-click, double-click, scroll, drag & drop via live screenshot |\n| ⌨️ **Keyboard control** | Text typing (Unicode/Turkish included, layout-independent), keys and shortcuts (Ctrl+C, Alt+Tab…) |\n| 👁️ **OCR** | Converts on-screen text to machine-readable format |\n| 📷 **Vision access** | Raw-pixel paths for image-capable models: single frames, MJPEG stream, text-based motion detection |\n| 🪟 **Window management** | List, focus, safe close (WM_CLOSE), kill (task-manager style) |\n| 🖥️ **Focus-free control** | Read/write background windows via PostMessage without stealing focus |\n| 🎮 **Game mode** | Camera look via relative mouse movement, hold-to-move keys |\n| 🔐 **Token auth** | Every request requires `X-Auth-Token` (CSRF protection) |\n| 🦺 **Stuck-input watchdog** | Auto-releases held keys after 30 s of inactivity |\n| 🛟 **Failsafe** | Cursor to top-left corner aborts all commands (disabled in game mode) |\n\n---\n\n## Architecture\n\n```\n┌─────────────────────────────────────────────────────────┐\n│ Browser (Web UI) │\n│ ┌──────────┐ ┌──────────┐ ┌────────────────────────┐ │\n│ │ Live │ │ Control │ │ Windows / Game Mode │ │\n│ │ View │ │ Panel │ │ Panel │ │\n│ └────┬─────┘ └────┬─────┘ └───────────┬────────────┘ │\n│ │ │ │ │\n└───────┼──────────────┼─────────────────────┼──────────────┘\n │ │ │\n ▼ ▼ ▼\n┌─────────────────────────────────────────────────────────┐\n│ HTTP API (Flask) │\n│ 127.0.0.1:8745 │\n│ │\n│ /api/screenshot /api/mouse /api/key │\n│ /api/vision/* /api/ocr /api/window │\n│ /api/game /api/held /api/release_all │\n│ /api/windows /api/desktops /api/desktop │\n│ │\n│ ┌─────────────┐ ┌──────────────┐ ┌──────────────┐ │\n│ │ Auth Layer │ │ Watchdog │ │ OCR Engine │ │\n│ │ (token) │ │ (30s auto) │ │ (RapidOCR) │ │\n│ └─────────────┘ └──────────────┘ └──────────────┘ │\n└─────────────────────────────────────────────────────────┘\n │ │ │\n ▼ ▼ ▼\n┌─────────────────────────────────────────────────────────┐\n│ control.py (Core) │\n│ │\n│ Screen: mss (fast capture), PIL (processing) │\n│ Mouse: pyautogui (absolute), SendInput (relative) │\n│ Keyboard: pyautogui + SendInput+KEYEVENTF_UNICODE │\n│ Windows: Win32 API (EnumWindows, SetForegroundWindow) │\n│ Background: PrintWindow (capture), PostMessage (input) │\n│ Virtual Desktops: pyvda │\n│ Game Mode: ClipCursor + MOUSE_MOVE_RELATIVE │\n└─────────────────────────────────────────────────────────┘\n┌──────────────────────────────────────────────┐\n│ backends/ (pluggable) │\n│ WindowsBackend │ LinuxBackend │ MacOSBackend│\n│ (full) │ (stub) │ (stub) │\n└──────────────────────────────────────────────┘\n```\n\n### Coordinates & Concurrency\n\n**Per-Monitor DPI awareness.** `control.py` calls\n`SetProcessDpiAwarenessContext(PER_MONITOR_AWARE_V2)` at import time —\n**before** the `pyautogui` import, because pyautogui touches coordinate APIs\nduring import and would otherwise lock the process to the interpreter\nmanifest's default (system-aware). With PMv2 active, every coordinate in the\nsystem is a **physical pixel** end to end: mss capture, OCR bounding boxes,\npyautogui/SendInput clicks, ClipCursor. On High-DPI displays (125%/150%\nscaling) nothing drifts between what OCR reports and where the mouse clicks.\n\n**Lock architecture.** The server uses two independent locks instead of one\nglobal lock:\n\n| Lock | Protects | Endpoints |\n|---|---|---|\n| `_input_lock` | mouse, keyboard, game mode, window ops | `/api/mouse`, `/api/key`, `/api/game`, `/api/window/post`, ... |\n| `_read_lock` | capture, OCR, vision, enumeration | `/api/screenshot`, `/api/ocr`, `/api/vision/*`, `/api/windows`, ... |\n\nA slow OCR (3–5 s on a busy screen) no longer freezes concurrent screenshot\nor vision reads — reads queue behind reads, inputs behind inputs.\n\n### Live-Loop Working Principle\n\nThis system is designed for a **live perceive-act loop**, not pre-written\ncommand chains:\n\n1. **READ** — OCR or vision reads the screen before and after every action\n2. **ONE ACTION** — each round sends a single command\n3. **VERIFY** — acceptance is \"it appeared on screen\", not \"I sent it\"\n4. **ADAPT** — if verification fails, the next step changes based on what\n is actually seen\n\nThis is enforced by the `expect_hwnd` guard: typing is **refused (409)** if\nthe foreground window doesn't match the target.\n\n---\n\n## Installation\n\n```bash\ncd screen-control\npip install -r requirements.txt\n```\n\n### Requirements\n\n| Package | Purpose | Required? |\n|---|---|---|\n| `mss` | Fast screen capture | ✅ Yes |\n| `pyautogui` | Mouse/keyboard control | ✅ Yes |\n| `pyvda` | Virtual desktop management | ✅ Yes |\n| `flask` | HTTP server | ✅ Yes |\n| `Pillow` | Image processing | ✅ Yes |\n| `rapidocr-onnxruntime` | OCR (screen text reading) | ⚠️ Optional |\n\n> **Note:** The OCR package is large and may take a while to install. If it\n> fails, everything else still works — only the OCR feature is unavailable.\n\n### System Requirements\n\n- **OS:** Windows 10/11 (x64)\n- **Python:** 3.10+\n- **Display:** Any resolution; the system adapts automatically\n\n---\n\n## Quick Start\n\n```bash\n# 1. Start the server\ncd screen-control\npython server.py\n\n# 2. Open in browser\n# http://127.0.0.1:8745\n\n# 3. Or control via API\nTOKEN=$(cat .token)\ncurl -H \"X-Auth-Token: $TOKEN\" http://127.0.0.1:8745/api/screenshot -o screen.jpg\n```\n\n---\n\n## API Reference\n\n### Capabilities\n\nReturns the active backend name and what it can do. Agents should call this\nfirst (see [ROADMAP.md](https://github.com/Xeakaes/computer-use-for-all-agents/blob/main/ROADMAP.md) for the multi-platform plan).\n\n```http\nGET /api/capabilities\n→ {\"ok\": true, \"backend\": \"windows\",\n \"capabilities\": {\"screen_capture\": true, \"game_mode\": true, ...}}\n```\n\nFeature values: `true` (supported), `false` (absent), `null` (unknown —\nstub backend), `\"optional\"` (depends on an optional dependency).\n\n### Platform Support Matrix\n\n| Capability | Windows | Linux X11 | Linux Wayland | macOS |\n|---|---:|---:|---:|---:|\n| Screen capture | Full | Full | Portal-dependent | Permission required |\n| OCR | Full/optional | Full/optional | Full/optional | Full/optional |\n| Mouse control | Full | Full | Restricted | Accessibility permission |\n| Keyboard control | Full | Full | Restricted | Accessibility permission |\n| Window enumeration | Full | WM-dependent | Limited | Accessibility/API-dependent |\n| Background input | Strong | WM/app-dependent | Usually unavailable | Limited |\n| Virtual desktops | Supported | DE/WM-dependent | DE/WM-dependent | Spaces-specific |\n| Game mode | Supported | Experimental | Limited | Experimental |\n\n> Linux and macOS backends are currently **fail-closed stubs**: every\n> operation returns `BACKEND_UNAVAILABLE` (501) until implemented\n> (ROADMAP Phases 5–7). Windows is the reference backend.\n\n### Authentication\n\nEvery request **must** include the `X-Auth-Token` header. The token is\ngenerated on each server start and written to `.token`.\n\n```bash\nTOKEN=$(cat .token)\n```\n\n| Code | Meaning |\n|---|---|\n| 401 | Missing or invalid token |\n| 415 | POST without `Content-Type: application/json` |\n\n**Token bootstrap** (for the bundled web UI):\n\n```http\nGET /token\n→ {\"ok\": true, \"token\": \"abc123...\"}\n```\n\n> The `/token` endpoint is safe: Same-Origin Policy prevents foreign pages\n> from reading it.\n\n---\n\n### Screen Capture\n\n#### `GET /api/screenshot`\n\nReturns a JPEG screenshot.\n\n| Parameter | Type | Default | Description |\n|---|---|---|---|\n| `monitor` | int | 1 | Monitor index |\n| `region` | string | — | `x,y,w,h` sub-region |\n\n```bash\ncurl -H \"X-Auth-Token: $TOKEN\" -o screen.jpg http://127.0.0.1:8745/api/screenshot\ncurl -H \"X-Auth-Token: $TOKEN\" \"http://127.0.0.1:8745/api/screenshot?region=0,0,800,600\"\n```\n\n#### `GET /api/info`\n\nReturns screen dimensions and system state.\n\n```json\n{\"ok\": true, \"width\": 1920, \"height\": 1080, \"ocr_available\": true,\n \"failsafe\": true, \"game_mode\": false}\n```\n\n---\n\n### Mouse Control\n\n#### `POST /api/mouse`\n\n| action | Required params | Optional params | Description |\n|---|---|---|---|\n| `move` | `x`, `y` | `duration` (default 0.15) | Move cursor to absolute position |\n| `click` | `x`, `y` | `button` (left/right), `clicks` (default 1) | Click at position |\n| `scroll` | `clicks` | `x`, `y` | Scroll wheel (positive=up) |\n| `drag` | `x1`, `y1`, `x`, `y` | `duration`, `button` | Drag between two points |\n| `down` | `button` (default \"left\") | — | Press and hold mouse button |\n| `up` | `button` (default \"left\") | — | Release held mouse button |\n\n```bash\n# Click at center of screen\ncurl -X POST http://127.0.0.1:8745/api/mouse -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"action\":\"click\",\"x\":960,\"y\":540,\"button\":\"left\"}'\n\n# Right-click\ncurl -X POST http://127.0.0.1:8745/api/mouse -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"action\":\"click\",\"button\":\"right\",\"x\":960,\"y\":540}'\n\n# Scroll down\ncurl -X POST http://127.0.0.1:8745/api/mouse -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"action\":\"scroll\",\"clicks\":-3}'\n```\n\n---\n\n### Keyboard Control\n\n#### `POST /api/key`\n\n| action | Required params | Description |\n|---|---|---|\n| `press` | `key` | Press and release a key |\n| `down` | `key` | Hold a key down (tracked for watchdog) |\n| `up` | `key` | Release a held key |\n| `hotkey` | `keys` (array) | Key combination (e.g. `[\"ctrl\",\"c\"]`) |\n| `type` | `text` | Type text (Unicode, layout-independent) |\n\n| Optional param | Default | Description |\n|---|---|---|\n| `expect_hwnd` | — | Window handle to verify focus (409 if mismatch) |\n| `interval` | 0.03 | Delay between characters for `type` |\n\n```bash\n# Press Enter\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"press\",\"key\":\"enter\"}'\n\n# Ctrl+C\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"hotkey\",\"keys\":[\"ctrl\",\"c\"]}'\n\n# Type text (Turkish characters supported)\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"type\",\"text\":\"Merhaba dünya\"}'\n\n# Hold W key down (for walking in games)\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"down\",\"key\":\"w\"}'\n```\n\n---\n\n### OCR (Screen Reading)\n\n#### `POST /api/ocr`\n\nConverts on-screen text to machine-readable format.\n\n| Param | Type | Default | Description |\n|---|---|---|---|\n| `region` | array | — | `[x, y, w, h]` sub-region (faster) |\n\n```json\n{\n \"ok\": true,\n \"text\": \"Hello World\\nFile Edit View\",\n \"lines\": [\"Hello World\", \"File Edit View\"],\n \"items\": [\n {\"text\": \"Hello World\", \"x\": 960, \"y\": 40},\n {\"text\": \"File Edit View\", \"x\": 100, \"y\": 15}\n ]\n}\n```\n\n```bash\n# Full screen OCR\ncurl -X POST http://127.0.0.1:8745/api/ocr -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{}'\n\n# Region-only (faster, ~10x for small regions)\ncurl -X POST http://127.0.0.1:8745/api/ocr -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"region\":[0,0,800,100]}'\n```\n\n---\n\n### Vision Access (Image Models)\n\nThree endpoints for models that can consume images:\n\n| Endpoint | Description |\n|---|---|\n| `GET /api/vision/frame` | Single JPEG frame (raw or base64) |\n| `GET /api/stream` | MJPEG live stream |\n| `POST /api/vision/diff` | Text-based motion detection (no vision needed) |\n\n#### `GET /api/vision/frame`\n\n| Param | Default | Description |\n|---|---|---|\n| `scale` | 1.0 | Downscale factor (0.5 = half size) |\n| `gray` | 0 | 1 for greyscale |\n| `quality` | 80 | JPEG quality (20-95) |\n| `format` | — | `base64` for JSON response |\n| `region` | — | `x,y,w,h` sub-region |\n\n```bash\n# Half-size greyscale frame as base64 (for text-only models)\ncurl \"http://127.0.0.1:8745/api/vision/frame?scale=0.5&gray=1&format=base64\" \\\n -H \"X-Auth-Token: $TOKEN\"\n```\n\n#### `GET /api/stream`\n\nMJPEG live stream. Drop into `` or consume frame-by-frame.\n\n| Param | Default | Description |\n|---|---|---|\n| `fps` | 10 | Frames per second (1-30) |\n| `quality` | 70 | JPEG quality |\n| `scale` | 1.0 | Downscale factor |\n| `region` | — | `x,y,w,h` sub-region |\n\n#### `POST /api/vision/diff`\n\nText-based motion detection — **no vision model required**.\n\n| Body | Description |\n|---|---|\n| `{}` | Compare against last stored frame |\n| `{\"grab\":\"gray\"}` | Store current frame for next comparison |\n| `{\"b64_prev\":\"...\"}` | Compare against provided previous frame |\n\n```json\n{\n \"ok\": true,\n \"changed\": true,\n \"changed_pct\": 12.5,\n \"bbox\": [100, 200, 400, 350],\n \"tiles\": [\n {\"row\": 2, \"col\": 4, \"pct\": 35.2, \"center\": [1000, 390]}\n ]\n}\n```\n\n---\n\n### Window Management\n\n#### `GET /api/windows`\n\nList all visible windows.\n\n```json\n{\n \"ok\": true,\n \"windows\": [\n {\n \"hwnd\": 123456,\n \"title\": \"My Application\",\n \"process\": \"app.exe\",\n \"pid\": 7890,\n \"focused\": true,\n \"rect\": [0, 0, 1920, 1080],\n \"desktop\": 1\n }\n ]\n}\n```\n\n#### `POST /api/window`\n\n| action | Required | Optional | Description |\n|---|---|---|---|\n| `focus` | `hwnd` | — | Bring window to foreground |\n| `close` | `hwnd` | `expect_title`, `expect_process` | Safe close via WM_CLOSE |\n| `kill` | `hwnd`, `pid` | — | Force kill (task-manager style) |\n| `topmost` | `hwnd` | — | Set always-on-top |\n| `untopmost` | `hwnd` | — | Remove always-on-top |\n| `maximize` | `hwnd` | — | Maximise window |\n\n```bash\n# Focus a window\ncurl -X POST http://127.0.0.1:8745/api/window -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"hwnd\":12345,\"action\":\"focus\"}'\n\n# Safe close (with title verification)\ncurl -X POST http://127.0.0.1:8745/api/window -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"action\":\"close\",\"hwnd\":12345,\"expect_title\":\"Notepad\"}'\n\n# Kill process\ncurl -X POST http://127.0.0.1:8745/api/window -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"kill\",\"hwnd\":12345,\"pid\":7890}'\n```\n\n---\n\n### Focus-Free (Background) Control\n\nRead and control windows **without stealing focus** — the user keeps working\non their main desktop.\n\n#### `GET /api/window/capture`\n\nCapture a window via PrintWindow (works even on another virtual desktop).\n\n| Param | Description |\n|---|---|\n| `hwnd` (required) | Window handle |\n| `client` | 1 = client area only |\n| `ocr` | 1 = return OCR text instead of image |\n\n```bash\n# Capture window as PNG\ncurl \"http://127.0.0.1:8745/api/window/capture?hwnd=12345\" \\\n -H \"X-Auth-Token: $TOKEN\" -o window.png\n\n# Capture + OCR in one call\ncurl \"http://127.0.0.1:8745/api/window/capture?hwnd=12345&ocr=1\" \\\n -H \"X-Auth-Token: $TOKEN\"\n```\n\n#### `POST /api/window/post`\n\nSend input to a window, choosing the delivery path automatically.\n\n| action | Description |\n|---|---|\n| `type` | Type text (Unicode-safe) |\n| `key` | Send a key press |\n| `hotkey` | Send a key combination |\n| `click` | Click at client coordinates |\n| `scroll` | Scroll the window |\n| `drag` | Drag inside the window |\n\nOptional `mode` parameter controls routing:\n\n| mode | Behavior |\n|---|---|\n| `auto` (default) | Decided by input-mode probe (see below) |\n| `background` | Force PostMessage path (window keeps focus/z-order) |\n| `focused` | Force focus + SendInput path |\n\n```bash\n# Type into a background Notepad\ncurl -X POST http://127.0.0.1:8745/api/window/post -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"hwnd\":12345,\"action\":\"type\",\"text\":\"Hello from background!\"}'\n```\n\nRouting rules (`mode=auto`):\n\n- **`postmessage`** — classic Win32 app: background PostMessage, no focus change.\n- **`uia`** — WinUI/UWP/XAML surface (single DirectX canvas, no Win32 child\n controls): posted messages are silently swallowed, so the window is focused\n and the action is replayed through SendInput (client coords converted to\n screen). This is the documented fallback for modern apps.\n- **`focused`** — window is already foreground: focused SendInput path.\n- **`invalid`** — HTTP 409; not a reachable top-level window.\n\n#### `GET /api/window/input-mode`\n\nClassify how a window receives input **before** posting to it. Returns one of\n`focused` | `postmessage` | `uia` | `invalid`.\n\n```bash\ncurl \"http://127.0.0.1:8745/api/window/input-mode?hwnd=12345\" \\\n -H \"X-Auth-Token: $TOKEN\"\n```\n\n> **WinUI note:** New Notepad (and other XAML-hosted apps) has no classic\n> child Edit control to post to — the whole UI is one DirectX surface.\n> `input-mode` reports `uia` for these; `/api/window/post` then automatically\n> uses the focused SendInput path. `/api/window/children` remains useful for\n> classic apps with real child controls.\n\n#### `GET /api/window/children`\n\nList child controls of a window (class name + title + hwnd).\n\n```bash\ncurl \"http://127.0.0.1:8745/api/window/children?hwnd=12345\" -H \"X-Auth-Token: $TOKEN\"\n```\n\n---\n\n### Virtual Desktops\n\n#### `GET /api/desktops`\n\nList all virtual desktops.\n\n#### `POST /api/desktop`\n\n| action | Params | Description |\n|---|---|---|\n| `switch` | `number` | Switch to desktop N |\n| `create` | — | Create a new desktop |\n\n```bash\ncurl http://127.0.0.1:8745/api/desktops -H \"X-Auth-Token: $TOKEN\"\n\ncurl -X POST http://127.0.0.1:8745/api/desktop -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"switch\",\"number\":2}'\n```\n\n---\n\n### Game Mode\n\n| action | Params | Description |\n|---|---|---|\n| `start` | `sensitivity` (default 12) | Lock cursor to center, enable game input |\n| `move` | `dx`, `dy`, `sensitivity` | Rotate camera (relative mouse) |\n| `stop` | — | Release cursor + all held input |\n| `heartbeat` | — | Keep-alive for long holds |\n\n```bash\n# Start game mode\ncurl -X POST http://127.0.0.1:8745/api/game -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"start\",\"sensitivity\":12}'\n\n# Look right\ncurl -X POST http://127.0.0.1:8745/api/game -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"move\",\"dx\":50,\"dy\":0}'\n\n# Hold W to walk forward\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"down\",\"key\":\"w\"}'\n\n# ... later ...\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"up\",\"key\":\"w\"}'\n\n# Stop game mode\ncurl -X POST http://127.0.0.1:8745/api/game -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"stop\"}'\n```\n\n---\n\n### Safety Endpoints\n\n#### `GET /api/held`\n\nReturns currently held keys/buttons and watchdog status.\n\n```json\n{\n \"ok\": true,\n \"keys\": [\"w\", \"shift\"],\n \"buttons\": [\"left\"],\n \"game_mode\": true,\n \"idle_seconds\": 5.2,\n \"watchdog_count\": 0,\n \"last_watchdog\": null\n}\n```\n\n#### `POST /api/release_all`\n\nEmergency: release **everything** (held keys, mouse buttons, game-mode cursor\nlock).\n\n```bash\ncurl -X POST http://127.0.0.1:8745/api/release_all -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{}'\n```\n\n---\n\n---\n\n## 🤖 For AI Agents\n\nA dedicated, comprehensive guide for AI agents (LLMs, vision models, automation\nframeworks) is available in **[AGENT_GUIDE.md](https://github.com/Xeakaes/computer-use-for-all-agents/blob/main/AGENT_GUIDE.md)**.\n\nIt covers:\n- Perceive-act loop (read → plan → act → verify)\n- Focus guard (`expect_hwnd`) to prevent wrong-window accidents\n- App automation and game control workflows\n- Vision access for image-capable models\n- Text-based motion detection\n- Bandwidth optimization\n- Complete curl examples\n\n---\n\n## MCP Support (One-Click Cloud Agents)\n\n**Model Context Protocol (MCP)** turns this project into a plug-and-play\ntoolbox for any MCP-capable agent: Claude Desktop, Claude Code, Cursor,\nVS Code Copilot Agent mode, custom cloud agents — no custom glue code,\nno curl scripts. The agent discovers and calls the tools natively.\n\n### How it works\n\n```\nMCP agent (cloud or desktop)\n │ MCP protocol (stdio or streamable-HTTP)\n ▼\n mcp_server.py ← thin wrapper: tools → HTTP calls, token auto-read\n │ REST + X-Auth-Token (localhost only)\n ▼\n server.py ← the single source of truth:\n auth, locks, watchdog, focus guard, all safety rules\n```\n\n`mcp_server.py` adds **no new powers** — every safety mechanism\n(auth token, input/read locks, watchdog, Alt+F4 block, focus guard,\nfailsafe) stays enforced by `server.py`.\n\n### Setup\n\n```bash\npip install mcp # optional dependency (see requirements.txt)\npython server.py # start the REST server first (it writes .token)\n```\n\nThe MCP server auto-reads the token from `.token` (or the\n`SCREEN_CONTROL_TOKEN` env var) — zero configuration.\n\n### Desktop agents (stdio transport)\n\nClaude Desktop — `claude_desktop_config.json`:\n\n```json\n{\n \"mcpServers\": {\n \"screen-control\": {\n \"command\": \"python\",\n \"args\": [\"C:/path/to/screen-control/mcp_server.py\"]\n }\n }\n}\n```\n\nClaude Code: `claude mcp add screen-control -- python C:/path/to/screen-control/mcp_server.py`\n\nCursor / VS Code: add the same entry to their MCP config files.\n\n### Remote / cloud agents (streamable-HTTP transport)\n\n```bash\npython mcp_server.py --http --port 8751\n# MCP endpoint: http://127.0.0.1:8751/mcp\n```\n\nThe HTTP transport is **token-protected**: every request must carry the\n`X-Auth-Token` header (same token as the REST server). Query-string tokens\n(`?token=...`) are **rejected by design** — URLs leak into proxy/tunnel\nlogs, browser history and shared links, and this token grants full desktop\ncontrol. Clients that cannot send custom headers should run a local stdio\n`mcp_server.py` instead. Only `GET /health` is open, for liveness probes.\nDNS-rebinding protection is disabled on this transport deliberately —\ntunneled requests arrive with a foreign `Host` header, and the rebinding\nthreat is already covered by the token guard.\n\nFor a cloud agent, expose it through a tunnel:\n\n```bash\ncloudflared tunnel --url http://127.0.0.1:8751\n# → prints a https://.trycloudflare.com URL\n```\n\nThen configure the agent's MCP connection with `/mcp` plus the\ntoken from `.token` as a header (`X-Auth-Token`).\n\n### Headerless connectors (scoped keys)\n\nFor clients that cannot send custom headers (e.g. web connectors that only\ntake an endpoint URL), create a **scoped API key** — a persistent, optionally\ntime-limited credential — and embed it in the URL path:\n\n```bash\n# Create a 24-hour scoped key (requires the server to be running)\ncurl -X POST http://127.0.0.1:8745/api/keys -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"action\":\"create\",\"name\":\"spark\",\"expires_in_hours\":24}'\n\n# Connector endpoint becomes:\n# https:///mcp/\n```\n\nDesign guarantees (SC-06):\n- The **master session token is refused in URLs** (403) — only scoped keys\n may travel there\n- Scoped keys **expire automatically**; expired keys authenticate nothing\n- Scoped keys are **revocable by name** at any moment via `POST /api/keys`\n (`{\"action\":\"revoke\",\"name\":\"spark\"}`) — revocation takes effect\n immediately on every endpoint\n- The `.apikeys` file stores only SHA-256 hashes, never raw keys\n\n> ⚠️ A tunnel exposes PC control to the internet. Keep the token secret,\n> prefer short-lived tunnels and scoped keys for headerless connectors,\n> and stop the server when not in use.\n\n### One-Command Startup (launcher + auto-tunnel)\n\n`start-server.bat` automates the whole cloud setup and prints everything\nyour cloud agent needs, ready to paste:\n\n1. Downloads `cloudflared.exe` if missing (portable, no admin required)\n2. Stops leftover instances from a previous run\n3. Starts the REST server (port 8745) and the MCP HTTP server (port 8751)\n4. Waits until both are healthy (`/token` and `/health` probes)\n5. Starts a cloudflared quick tunnel, extracts its public URL from\n `tunnel.log`, and prints the summary:\n\n```\n ============================================================\n ALL SYSTEMS RUNNING\n ============================================================\n Local REST API : http://127.0.0.1:8745\n Local MCP : http://127.0.0.1:8751/mcp\n Public MCP URL : https://.trycloudflare.com/mcp\n\n ------------------------------------------------------------\n PASTE INTO YOUR CLOUD AGENT (MCP connector settings)\n ------------------------------------------------------------\n Endpoint : https://.trycloudflare.com/mcp\n Header : X-Auth-Token: \n URL form : https://.trycloudflare.com/mcp?token=\n (only if the connector cannot send headers)\n ------------------------------------------------------------\n```\n\n`stop-server.bat` stops all three (REST, MCP, tunnel) in one go.\n\n### Available tools (16)\n\n| Category | Tools |\n|---|---|\n| Perception | `get_info`, `ocr_screen`, `screenshot` (real image block for vision models), `motion_diff` |\n| Mouse / keyboard | `mouse`, `keyboard` (with `expect_hwnd`), `get_held`, `release_all` |\n| Windows | `list_windows`, `focus_window`, `window_children`, `window_input_mode`, `window_post`, `window_capture_ocr`, `close_window` |\n| Game mode | `game` (start / move / stop / heartbeat) |\n\n### Which transport for whom\n\n| Consumer | Transport | Command |\n|---|---|---|\n| Claude Desktop / Cursor / VS Code (local) | stdio | `python mcp_server.py` |\n| Claude Code | stdio | `claude mcp add ...` (above) |\n| Cloud / remote agents | streamable-HTTP | `start-server.bat` (recommended) or `python mcp_server.py --http --port 8751` + `cloudflared tunnel --url http://127.0.0.1:8751` |\n\n> **Note:** This project targets MCP Python SDK 2.x (`MCPServer` API).\n> With SDK 1.x, replace the import with `from mcp.server.fastmcp import FastMCP, Image`\n> and `MCPServer` with `FastMCP`.\n\n---\n\n## Security Model\n\n### Threat: Malicious Web Pages (CSRF)\n\nEven bound to `127.0.0.1`, a malicious page in the browser can trigger\nnon-preflighted requests (text/plain fetch, HTML form POST) to localhost.\nThe browser blocks the **response** but not the **request** — the server\nwould still execute the command.\n\n**Mitigation:** Every request requires `X-Auth-Token`. A foreign page\ncannot read this token (Same-Origin Policy), so it cannot authenticate.\n\nAdditional layers:\n- POST requests **must** use `Content-Type: application/json` (415 otherwise)\n- This blocks form-encoded and text-plain POSTs even if the token leaked\n- **Host header trust (DNS rebinding):** when bound to loopback, requests\n carrying a non-loopback `Host` header are refused with `421` — a\n rebinding page that resolves its domain to `127.0.0.1` cannot read\n `/token` or call the API\n- `/token` and `/` responses carry `Cache-Control: no-store` so the\n credential is never persisted by browsers or proxies\n\n### Threat: Stuck Keys / Game Mode Lock\n\nIn game mode, `ClipCursor` pins the cursor to a 2×2 box — the classic\npyautogui failsafe (cursor to top-left) **does not work**.\n\n**Mitigations:**\n1. **Physical `Esc` / `Alt+Tab`** — real hardware input; this API cannot\n block it, and it always works\n2. `POST /api/release_all` — instant release of everything\n3. **Watchdog (automatic)** — 30 s of server-side inactivity with held input\n triggers automatic release\n\n### Threat: Wrong Window Typing\n\n**Mitigations:**\n- `expect_hwnd` guard on `/api/key` — if the foreground window doesn't\n match, typing is refused with 409\n- The focused path of `/api/window/post` verifies the focus **after** the\n focus switch and **before** any synthetic input (409 on mismatch) —\n input is never replayed into whatever window happens to be foreground\n- `focus_window()` raises on failure instead of silently returning\n\n### Threat: Dangerous Key Combos\n\n**Mitigation:** Blocked at the API level (403) on **every delivery path** —\nthe direct `/api/key` route, the background `/api/window/post` route\n(PostMessage), and the focused fallback route share one safety policy\n(`control._assert_allowed`):\n- `Alt+F4` — the only banned Alt combo (Alt+Tab, Alt+menu are legitimate)\n- Win key — prevents Start menu, task switching\n- `Ctrl+Alt+Del` — system security screen\n- `Shift+Delete` style — prevents permanent deletion\n\n### Threat: Killing System Processes\n\n**Mitigations:**\n- The process name is resolved **from the PID directly** (Win32 toolhelp\n snapshot), not from the visible-window inventory — windowless/background\n system processes get the same protection as visible ones\n- Critical system processes are blacklisted (default-deny for unknown PIDs):\n `winlogon.exe`, `csrss.exe`, `smss.exe`, `services.exe`, `lsass.exe`,\n `svchost.exe`, `system`, `registry`, `dwm.exe`\n- Optional `expect_process` confirmation: a mismatch aborts the kill with\n `409` — protects against killing a newly-reused PID\n\n### Threat: Resource Exhaustion (rogue agent / DoS)\n\nA token-holding but misbehaving client should not be able to exhaust memory\nor starve the input lock.\n\n**Mitigations:**\n- `MAX_CONTENT_LENGTH` = 1 MB — oversized request bodies are rejected (413)\n- `region` width/height/area and `scale` are bounded (400 otherwise)\n- `text` payloads are capped at 10,000 characters per input call\n- MJPEG streams are capped at 10 concurrent clients (429 beyond that)\n- Pillow decompression-bomb limit is set for client-supplied images\n\n### Network Access\n\nThe server binds to `127.0.0.1` by default. To expose it to the network:\n\n```bash\npython server.py --host 0.0.0.0 # ⚠️ anyone on the network can control this machine\n```\n\n---\n\n## Game Mode Guide\n\n### Setup\n\n```bash\n# 1. Focus the game window\ncurl -X POST http://127.0.0.1:8745/api/window -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"hwnd\":GAME_HWND,\"action\":\"focus\"}'\n\n# 2. Start game mode\ncurl -X POST http://127.0.0.1:8745/api/game -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"start\",\"sensitivity\":12}'\n```\n\n### Camera Look\n\n```bash\n# Look right\ncurl -X POST http://127.0.0.1:8745/api/game -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"move\",\"dx\":50,\"dy\":0}'\n\n# Look down\ncurl -X POST http://127.0.0.1:8745/api/game -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"move\",\"dx\":0,\"dy\":30}'\n```\n\n### Movement\n\n```bash\n# Walk forward (hold W)\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"down\",\"key\":\"w\"}'\n\n# ... walk for a while ...\n\n# Release W\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"up\",\"key\":\"w\"}'\n```\n\n### Minecraft-Specific\n\n```bash\n# Place block (right-click)\ncurl -X POST http://127.0.0.1:8745/api/mouse -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" \\\n -d '{\"action\":\"click\",\"button\":\"right\",\"x\":960,\"y\":540}'\n\n# Break block (hold left-click)\ncurl -X POST http://127.0.0.1:8745/api/mouse -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"down\",\"button\":\"left\"}'\n\n# ... after breaking ...\n\ncurl -X POST http://127.0.0.1:8745/api/mouse -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"up\",\"button\":\"left\"}'\n\n# Select hotbar slot\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"press\",\"key\":\"1\"}'\n\n# Open inventory\ncurl -X POST http://127.0.0.1:8745/api/key -H \"X-Auth-Token: $TOKEN\" \\\n -H \"Content-Type: application/json\" -d '{\"action\":\"press\",\"key\":\"e\"}'\n```\n\n### Suitability\n\n| Game Type | Suitable? | Notes |\n|---|---|---|\n| Minecraft (building) | ✅ Yes | Place blocks, walk, mine |\n| Minecraft (PvP) | ❌ No | Too slow for fast combat |\n| Turn-based games | ✅ Yes | Ample time for read→act→verify |\n| RPG / adventure | ✅ Yes | Inventory, dialogue, exploration |\n| Fast FPS | ❌ No | Reaction time insufficient |\n| Puzzle games | ✅ Yes | Click-based, read-heavy |\n\n---\n\n## Vision Access Guide\n\n### For Image-Capable Models\n\nIf the consuming model can process images, use the vision endpoints directly:\n\n```\nGET /api/vision/frame?scale=0.5&gray=1&quality=70\n```\n\nThis returns a single JPEG that the model can analyze for:\n- Game HUD elements (health, mana, inventory)\n- On-screen text (menus, chat, tooltips)\n- Visual scene understanding (blocks, entities, terrain)\n\n### For Text-Only Models\n\nUse the diff endpoint for motion detection without vision:\n\n```\nPOST /api/vision/diff {\"grab\":\"gray\"} → first call: stores frame\nPOST /api/vision/diff → subsequent calls: returns diff\n```\n\nThe response tells you **where** things changed (tile coordinates) and **how\nmuch** (percentage), which is sufficient for:\n- Detecting that an action had an effect\n- Locating moving elements on screen\n- Tracking animation state changes\n\n### Bandwidth Optimization\n\n| Approach | Payload | Use Case |\n|---|---|---|\n| `scale=1.0, gray=0` | ~500 KB | Full detail |\n| `scale=0.5, gray=1` | ~50 KB | Good for most vision models |\n| `scale=0.25, gray=1` | ~10 KB | Maximum compression |\n| `diff` (text) | ~1 KB | Text-only agents |\n| `region=...` | Variable | Focus on specific area |\n\n---\n\n## Troubleshooting\n\n### \"OCR engine not installed\"\n\n```bash\npip install rapidocr-onnxruntime\n```\n\n### Server won't start (port in use)\n\n```bash\n# Find the process using port 8745\nnetstat -ano | findstr \":8745\"\n\n# Kill it\ntaskkill /PID /F\n```\n\n### \"Focus mismatch\" (409) when typing\n\nThe foreground window changed between the focus call and the type call.\nSolution: always pass `expect_hwnd` and verify focus before typing.\n\n### Window not found\n\nThe window may have been closed or may be a system window that\n`EnumWindows` doesn't expose. Try:\n```bash\ncurl http://127.0.0.1:8745/api/windows -H \"X-Auth-Token: $TOKEN\"\n```\n\n### Game mode cursor stuck\n\nUse `POST /api/release_all` or press `Esc` / `Alt+Tab` physically.\n\n### High OCR latency\n\nOCR on a full 1920×1080 screen can take from a few seconds up to ~30 s\ndepending on your CPU and on-screen complexity. Use a region — small crops\nare typically 10× faster:\n```json\n{\"region\": [0, 0, 800, 100]}\n```\n\n### Turkish characters not appearing\n\nThe system uses `SendInput + KEYEVENTF_UNICODE` which is layout-independent.\nIf characters still don't appear, the target app may not support Unicode\ninput — try `POST /api/window/post` with `action: \"type\"` instead.\n\n---\n\n## Project Structure\n\n```\nscreen-control/\n├── server.py # Flask HTTP server + all API endpoints\n├── core/ # PlatformBackend interface + standardized errors\n│ ├── backends.py # Abstract backend + lazy discovery\n│ └── errors.py # ApiError envelope + error codes\n├── backends/ # OS implementations behind PlatformBackend\n│ ├── windows.py # Reference backend (moved from control.py)\n│ ├── linux.py # Fail-closed stub (ROADMAP Phase 5)\n│ ├── macos.py # Fail-closed stub (ROADMAP Phase 7)\n│ ├── fake.py # In-memory backend for tests\n│ └── forbidden.py # Shared blocked-key policy\n├── control.py # Compatibility shim re-exporting backends.windows\n├── tests/unit/ # Offline unit + integration tests (no real input)\n├── mcp_server.py # MCP server (stdio + streamable-HTTP) — thin wrapper over the API\n├── sdk/\n│ └── screen_control.py # Python SDK client (pip-installable style)\n├── index.html # Bundled web UI (live view + control panels)\n├── docs/images/ # README assets (demo GIF captured by the API itself)\n├── requirements.txt # Python dependencies\n├── start-server.bat # One command: REST + MCP + cloud tunnel (Windows)\n├── stop-server.bat # Stop all three processes\n├── test-security.py # Security + game-mode test suite (34 checks)\n├── test-game.py # Live game-mechanics test (app launch → draw → safe close)\n├── test-endtoend.py # End-to-end test: open Notepad → type → save → verify\n├── .github/workflows/ # CI: runs the security suite on every push\n├── AGENT_GUIDE.md # AI agent integration guide (separate from this file)\n├── README.md # This file\n└── .token # Auto-generated auth token (gitignored)\n```\n\n---\n\n## Testing\n\n### Prerequisites\n\nThe server must be running:\n\n```bash\ncd screen-control\npython server.py\n```\n\n### Security test suite\n\nTests authentication, blocked key combos, window management, safe close,\ncritical process protection, and game mode — all non-destructive.\n\n```bash\ncd screen-control\npython test-security.py\n```\n\nExpected output:\n\n```\n== Token Authentication ==\n✓ Missing token -> 401\n✓ Wrong token -> 401\n✓ Correct token -> 200\n✓ Non-JSON POST -> 415\n\n== Blocked Key Combos ==\n✓ Alt+F4 blocked (403)\n✓ Win key blocked (403)\n✓ Win+D blocked (403)\n✓ Delete blocked (403)\n\n== Window List ==\n✓ Windows list requires GET\n✓ Window list is non-empty — 8 windows\n✓ Exactly one focused window\n\n== Safe Close Verification ==\n✓ Wrong title aborts close\n\n== Critical Process Protection ==\n✓ System process (pid 4) rejected (403)\n✓ pid 0 rejected (403)\n\n== Game Mode ==\n✓ Game mode started\n✓ Relative camera look\n✓ Game mode stopped\n\n== Watchdog (dry run) ==\n✓ Held state returns ok\n✓ Watchdog count reported\n\n========================================\nRESULT: 19 passed, 0 failed\n```\n\n### Live game-mechanics test\n\nLaunches a real application (mspaint or notepad), performs hold-to-draw\ngame mechanics, verifies via pixel analysis, then safely closes with\n\"Don't Save\" dialog handling.\n\n```bash\ncd screen-control\npython test-game.py\n```\n\n> **Note:** This test launches a real application. It handles cleanup\n> automatically (sends WM_CLOSE and clicks \"Don't Save\" if a dialog appears).\n\n---\n\n## Contributing\n\n1. Fork the repository\n2. Create a feature branch\n3. Make your changes\n4. Test on a Windows machine\n5. Submit a pull request\n\n### Code Style\n\n- **Python:** PEP 8, type hints, docstrings on all public functions\n- **Docstrings:** English, Google style\n- **Error messages:** English, descriptive\n- **Comments:** English, explain *why* not *what*\n\n---\n\n## License\n\nMIT License. See [LICENSE](https://github.com/Xeakaes/computer-use-for-all-agents/blob/main/LICENSE) for details.\n\n---\n\n*Built with ❤️ for local automation and AI agent research.*\n","changelogMarkdown","videoKey","requiredSecretsJson","publishedAt",1789675687000,"priceCents","trialDays","iconType","image","upstream",{"_111":112,"_113":114,"_115":116,"_117":118,"_119":68,"_120":121},"requiredSecrets",[],"tags",[98,99,100,101,102,103,104,105,106,107,108,109,110],"assets",[],"stats",{"_96":44,"_97":44},"ratingAvg","reviewCount","agent-tools","ai-agents","automation","computer-use","desktop-automation","gui-automation","llm-tools","ocr","python","remote-control","screen-capture","vision","windows","topics",[98,99,100,101,102,103,104,122,105,106,107,108,109,110],"license","MIT","language","Python","registryName","io.github.Xeakaes/computer-use-for-all-agents","homepage","packages",[],"mcp","ENV",{"_125":126,"_127":128,"_129":130,"_131":-7},"ENVIRONMENT","production","TURNSTILE_SITE_KEY","0x4AAAAAAB8-rEI2gNDf_QhU","SENTRY_DSN","https://45fd1229f78089d5c3ee78be128030e1@o4510583261757440.ingest.de.sentry.io/4510584253448272","SENTRY_ENABLE_LOCAL"]