{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"npm_npm-atomicbotaicomputer-use-mcp","slug":"npm-atomicbotaicomputer-use-mcp","name":"@atomicbotai/computer-use-mcp","type":"mcp","url":"https://www.npmjs.com/package/@atomicbotai/computer-use-mcp","page_url":"https://unfragile.ai/npm-atomicbotaicomputer-use-mcp","categories":["mcp-servers"],"tags":[],"pricing":{"model":"open_source","free":true,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_0","uri":"capability://tool.use.integration.desktop.automation.via.mcp.protocol","name":"desktop-automation-via-mcp-protocol","description":"Exposes desktop computer-use capabilities (mouse, keyboard, screen interaction) as standardized MCP tools that can be called by any MCP-compatible client. Implements the Model Context Protocol server pattern to translate high-level automation intents into low-level OS input events, enabling LLM agents to interact with GUI applications without native bindings or browser automation frameworks.","intents":["I want my LLM agent to click buttons, type text, and navigate UI elements on my desktop","I need to automate repetitive GUI tasks across multiple applications without writing Selenium or Playwright scripts","I want to expose desktop control to Claude or other MCP-aware models without building custom integrations"],"best_for":["LLM agent developers building autonomous desktop automation workflows","Teams integrating Claude or other MCP-compatible models with legacy GUI applications","Developers prototyping cross-application automation without learning application-specific APIs"],"limitations":["Limited to screen-based interaction — cannot directly access application state or APIs, only what's visible on screen","No built-in OCR or vision processing — relies on client to provide screen coordinates or text locations","Single-user, single-session model — concurrent desktop sessions not supported","No native support for multi-monitor setups or complex window management scenarios","Latency between action and screen update may cause race conditions in rapid-fire automation sequences"],"requires":["Node.js 16+ runtime","MCP-compatible client (Claude Desktop, custom MCP client, or LLM framework with MCP support)","Desktop environment with X11, Wayland, or Windows input APIs available","Appropriate OS-level permissions for input simulation (may require sudo on Linux/macOS)"],"input_types":["structured JSON commands (mouse coordinates, keyboard key names, text strings)","screen coordinates as integers","keyboard event specifications (key codes, modifiers)"],"output_types":["confirmation of action execution","screen capture/screenshot data","structured feedback on action success/failure"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_1","uri":"capability://automation.workflow.mouse.control.with.coordinate.targeting","name":"mouse-control-with-coordinate-targeting","description":"Provides granular mouse control through MCP tool calls that accept screen coordinates and execute movement, clicking (left/right/middle button), and drag operations. Translates coordinate-based commands into native OS input events using platform-specific APIs (xdotool on Linux, pyautogui-equivalent on Windows/macOS), with optional screen coordinate validation to prevent out-of-bounds clicks.","intents":["I need my agent to click specific UI elements identified by their screen coordinates","I want to perform drag-and-drop operations between two screen locations","I need to right-click context menus or perform multi-button mouse interactions"],"best_for":["Automation workflows targeting GUI applications with fixed or predictable layouts","Agents that receive screen coordinates from vision models or OCR systems","Cross-platform automation requiring consistent mouse behavior across Windows, macOS, and Linux"],"limitations":["Requires exact pixel coordinates — no built-in element detection or fuzzy matching","No hover state tracking — cannot detect or wait for hover-triggered UI changes","Drag operations may fail if target application doesn't support standard mouse drag events","No double-click or multi-click sequences — each click is a separate command","Screen coordinate system assumes single monitor or primary display only"],"requires":["Screen resolution and coordinate system known to client","Platform-specific input simulation library (xdotool, Windows API, or macOS Quartz)"],"input_types":["x, y integer coordinates","button identifier (left, right, middle)","duration for drag operations (milliseconds)"],"output_types":["boolean success confirmation","error message if coordinates out of bounds"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_2","uri":"capability://automation.workflow.keyboard.input.with.text.and.key.events","name":"keyboard-input-with-text-and-key-events","description":"Provides keyboard automation through MCP tools supporting both text input (typing strings character-by-character or as bulk input) and discrete key events (Enter, Tab, Escape, modifier keys). Handles keyboard state management (shift, ctrl, alt, cmd modifiers) and translates high-level key names into platform-specific key codes, supporting both ASCII text and special key sequences.","intents":["I need my agent to type text into form fields or search boxes","I want to send keyboard shortcuts (Ctrl+C, Cmd+V, Alt+Tab) to switch applications or trigger actions","I need to navigate UI using Tab, Enter, and arrow keys without mouse interaction"],"best_for":["Automation of text-heavy workflows (form filling, code editing, terminal interaction)","Keyboard-driven application automation (terminal tools, text editors, keyboard-shortcut-heavy UIs)","Agents that need to combine text input with modifier key sequences"],"limitations":["No keyboard state persistence — cannot track which keys are currently held down across multiple commands","Text input assumes single keyboard layout — no support for non-ASCII input methods or IME (Input Method Editor)","No key-repeat or hold-duration support — each key press is instantaneous","Modifier key combinations require explicit sequencing — cannot atomically press Ctrl+Shift+S as a single command","No keyboard event capture or listening — one-way output only, cannot detect what user types"],"requires":["Platform-specific keyboard event API (xdotool, Windows SendInput, macOS Quartz)","Keyboard layout configuration matching the target system"],"input_types":["text strings for bulk input","individual key names (Enter, Tab, Escape, ArrowUp, etc.)","modifier specifications (shift, ctrl, alt, cmd)"],"output_types":["confirmation of keypress execution","error if unsupported key name provided"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_3","uri":"capability://image.visual.screen.capture.and.visual.feedback","name":"screen-capture-and-visual-feedback","description":"Captures the current desktop screen state and returns it as image data (PNG, JPEG, or base64-encoded format) that can be fed back to vision models or displayed to users. Implements screenshot functionality at the OS level, supporting full-screen capture or region-based cropping, enabling agents to observe the result of previous actions and make decisions based on visual state.","intents":["I need my agent to see the current state of the desktop after performing an action","I want to capture a specific region of the screen for analysis by a vision model","I need to verify that a UI element appeared or changed before proceeding with the next action"],"best_for":["Agents implementing feedback loops (action → screenshot → analysis → next action)","Workflows requiring visual verification of automation success","Integration with vision models (Claude Vision, GPT-4V) for screen understanding"],"limitations":["Full-screen capture includes all windows and overlays — no selective window capture","Screenshot latency may cause timing issues if screen state changes rapidly","Large screenshots consume significant bandwidth and token usage when sent to vision models","No built-in OCR or text extraction — vision model must process image to extract text","Region cropping requires pre-calculated coordinates — no dynamic element-based cropping"],"requires":["Display server access (X11, Wayland on Linux; native APIs on Windows/macOS)","Sufficient disk/memory for temporary image storage","Vision model integration if using screenshots for automated analysis"],"input_types":["optional region coordinates (x1, y1, x2, y2) for cropping","optional format specification (png, jpeg, base64)"],"output_types":["image data (PNG/JPEG binary or base64-encoded string)","image dimensions and format metadata"],"categories":["image-visual","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_4","uri":"capability://tool.use.integration.mcp.protocol.server.implementation","name":"mcp-protocol-server-implementation","description":"Implements the Model Context Protocol (MCP) server specification, exposing desktop automation tools through a standardized JSON-RPC interface that any MCP-compatible client can invoke. Handles MCP protocol negotiation, tool schema definition, and request/response serialization, allowing the server to be discovered and used by Claude Desktop, custom LLM frameworks, or other MCP clients without custom integration code.","intents":["I want to expose desktop automation to Claude Desktop or other MCP-aware applications","I need to integrate desktop control into a custom LLM framework that supports MCP","I want to build a multi-tool agent that combines desktop automation with other MCP servers"],"best_for":["Developers integrating with Claude Desktop or other MCP-native applications","Teams building custom LLM frameworks that support MCP protocol","Agents requiring composition of multiple MCP servers (desktop + web + database tools)"],"limitations":["MCP protocol overhead adds ~50-100ms per tool invocation due to JSON serialization","No built-in authentication or access control — relies on client-side security","Single MCP server instance per desktop — no multi-user isolation","Tool schema must be pre-defined — no dynamic tool generation based on runtime state","No streaming support for long-running operations — all tool calls are request/response"],"requires":["MCP-compatible client (Claude Desktop, custom framework with MCP support)","Node.js 16+ for running the MCP server","Network connectivity if running server on different machine (stdio or HTTP transport)"],"input_types":["MCP tool call requests (JSON-RPC format)","tool parameters matching pre-defined schema"],"output_types":["MCP tool call responses (JSON-RPC format)","structured tool results with success/error status"],"categories":["tool-use-integration","automation-workflow"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_5","uri":"capability://automation.workflow.cross.platform.input.abstraction","name":"cross-platform-input-abstraction","description":"Abstracts platform-specific input APIs (xdotool on Linux, Windows SendInput API, macOS Quartz Events) behind a unified interface, translating generic input commands into platform-native calls. Detects the runtime OS and loads appropriate input drivers, handling platform-specific quirks (key code mappings, coordinate systems, event timing) transparently to the MCP client.","intents":["I want to write automation that works on Windows, macOS, and Linux without conditional code","I need consistent keyboard and mouse behavior across different operating systems","I want to deploy the same automation script to different machines without modification"],"best_for":["Cross-platform automation teams supporting multiple OS deployments","Developers building portable LLM agents that run on any desktop OS","Organizations with heterogeneous desktop environments (mixed Windows/Mac/Linux)"],"limitations":["Platform detection is static at startup — cannot handle OS changes or VM migration mid-session","Some OS-specific features may be unavailable on all platforms (e.g., Windows-only key codes)","Input timing and event delivery vary by OS — automation may require OS-specific delays","Coordinate systems differ slightly (DPI scaling on Windows, Retina displays on macOS) — may require calibration","No support for non-standard input devices (game controllers, touch screens, styluses)"],"requires":["Platform-specific input library installed (xdotool on Linux, native APIs on Windows/macOS)","OS detection and conditional library loading at runtime"],"input_types":["generic input commands (mouse_click, key_press, type_text)","platform-agnostic key names and coordinates"],"output_types":["confirmation of input execution","platform-specific error messages if input fails"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"npm_npm-atomicbotaicomputer-use-mcp__cap_6","uri":"capability://automation.workflow.stateless.action.execution.model","name":"stateless-action-execution-model","description":"Executes each desktop automation action (mouse click, key press, screenshot) as an independent, stateless operation without maintaining session state or action history. Each MCP tool call is processed atomically and immediately, with no implicit state carryover between calls, requiring clients to explicitly manage sequences and handle timing/synchronization.","intents":["I want to execute individual desktop actions without worrying about server state management","I need to build agents that can recover from failures by replaying actions from a known state","I want to parallelize multiple automation sequences without state conflicts"],"best_for":["Stateless automation workflows where each action is independent","Agents implementing explicit state management and action sequencing","Distributed automation scenarios where multiple clients may interact with the same desktop"],"limitations":["No implicit action history or undo capability — client must track all actions","No built-in synchronization between actions — rapid sequences may race (e.g., click before screen updates)","Client must handle timing and waits explicitly — no automatic retry or backoff","No session persistence — server restart loses all context (though desktop state remains)","Difficult to implement complex workflows requiring state (e.g., 'click if element exists, otherwise type')"],"requires":["Client-side state management and action sequencing logic","Explicit timing/delay handling between dependent actions"],"input_types":["individual action commands (mouse_click, key_press, screenshot)"],"output_types":["immediate action result (success/failure)","no implicit state or history"],"categories":["automation-workflow","tool-use-integration"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":27,"verified":false,"data_access_risk":"moderate","permissions":["Node.js 16+ runtime","MCP-compatible client (Claude Desktop, custom MCP client, or LLM framework with MCP support)","Desktop environment with X11, Wayland, or Windows input APIs available","Appropriate OS-level permissions for input simulation (may require sudo on Linux/macOS)","Screen resolution and coordinate system known to client","Platform-specific input simulation library (xdotool, Windows API, or macOS Quartz)","Platform-specific keyboard event API (xdotool, Windows SendInput, macOS Quartz)","Keyboard layout configuration matching the target system","Display server access (X11, Wayland on Linux; native APIs on Windows/macOS)","Sufficient disk/memory for temporary image storage"],"failure_modes":["Limited to screen-based interaction — cannot directly access application state or APIs, only what's visible on screen","No built-in OCR or vision processing — relies on client to provide screen coordinates or text locations","Single-user, single-session model — concurrent desktop sessions not supported","No native support for multi-monitor setups or complex window management scenarios","Latency between action and screen update may cause race conditions in rapid-fire automation sequences","Requires exact pixel coordinates — no built-in element detection or fuzzy matching","No hover state tracking — cannot detect or wait for hover-triggered UI changes","Drag operations may fail if target application doesn't support standard mouse drag events","No double-click or multi-click sequences — each click is a separate command","Screen coordinate system assumes single monitor or primary display only","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.05,"quality":0.24,"ecosystem":0.3,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.25,"quality":0.25,"ecosystem":0.15,"match_graph":0.23,"freshness":0.12}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:23.903Z","last_scraped_at":"2026-05-03T14:23:57.594Z","last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=npm-atomicbotaicomputer-use-mcp","compare_url":"https://unfragile.ai/compare?artifact=npm-atomicbotaicomputer-use-mcp"}},"signature":"5xsrkGWnu8yscfoj7MOpLqH1rKQPx9azn54Lf9XPH7hhqgbFZ5BOW6YnNPKOKcH2gU17HMSI/gtnrc1kA83hDg==","signedAt":"2026-06-22T04:21:09.472Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/npm-atomicbotaicomputer-use-mcp","artifact":"https://unfragile.ai/npm-atomicbotaicomputer-use-mcp","verify":"https://unfragile.ai/api/v1/verify?slug=npm-atomicbotaicomputer-use-mcp","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}