<img src="docs/image/logo.png" alt="QuickDesk Logo" width="120"> <h1 align="center">QuickDesk</h1> The First AI-Native Remote Desktop Built-in MCP Server · AI Agents Control Any Remote Computer · Open-Source & Free <a href="https://github.com/barry-ran/QuickDesk/actions/workflows/quickdesk-windows.yml"> <img src="https://github.com/barry-ran/QuickDesk/actions/workflows/quickdesk-windows.yml/badge.svg" alt="Windows Build"> </a> <a href="https://github.com/barry-ran/QuickDesk/actions/workflows/quickdesk-macos.yml"> <img src="https://github.com/barry-ran/QuickDesk/actions/workflows/quickdesk-macos.yml/badge.svg" alt="macOS Build"> </a> <a href="https://github.com/barry-ran/QuickDesk/releases"> <img src="https://img.shields.io/github/v/release/barry-ran/QuickDesk" alt="Release"> </a> <a href="https://github.com/barry-ran/QuickDesk/stargazers"> <img src="https://img.shields.io/github/stars/barry-ran/QuickDesk" alt="Stars"> </a> <a href="LICENSE"> <img src="https://img.shields.io/github/license/barry-ran/QuickDesk" alt="License"> </a> English | <a href="README_zh.md">中文</a> <a href="https://t.me/+utsmSUhNLc1iNTY1">Telegram</a> | <a href="https://github.com/barry-ran/QuickDesk/issues">Issues</a> | <a href="https://github.com/barry-ran/QuickDesk/releases">Download</a>

QuickDesk is the first AI-native remote desktop — an open-source, free application with a built-in MCP (Model Context Protocol) Server that lets any AI agent see and control remote computers.

While other remote desktop tools only serve human users, QuickDesk extends AI Computer Use from "local machine only" to "any remote machine in the world". Connect Claude, GPT, Cursor, or any MCP-compatible AI to QuickDesk, and it can screenshot, click, type, drag, and automate across remote desktops — just like a human operator, but faster and 24/7.

Built on Google Chromium Remoting technology (the engine behind Chrome Remote Desktop), QuickDesk delivers industrial-grade performance, stability, and security refined over 10+ years by Google.

Main Interface

Remote Desktop

AI Config

Why QuickDesk?

AI-First: The Only Remote Desktop with MCP Support → MCP Integration Guide

Built-in MCP Server: AI agents connect via standard MCP protocol — no plugins, no hacks, zero configuration
Dual Transport: stdio mode (AI client spawns process) or HTTP/SSE mode (QuickDesk hosts MCP server for multi-client access)
Full Computer Use Toolkit: 20+ MCP tools — screenshot, click, type, drag, scroll, hotkeys, clipboard, and more
Real-Time Visibility: AI operations are displayed in the QuickDesk GUI in real time — the user sees every mouse move and keystroke, and can intervene at any time
Multi-Device AI Orchestration: AI can connect to and control multiple remote machines simultaneously — batch automation, cross-device workflows, fleet management
Guided Prompts: 9 built-in MCP prompt templates covering remote operation, server health check, batch automation, system diagnosis, screen analysis, multi-device orchestration, and SOP documentation

High-Performance Foundation

Open-Source & Free: MIT License, no feature restrictions, no connection limits, commercial use welcome
Self-Hosted: Deploy your own signaling and TURN relay servers, keep full control of your data
Commercial-Grade Foundation: Built on Chromium Remoting — the same technology powering Chrome Remote Desktop, refined by Google for 10+ years and proven by billions of users
Pure C++ Ultimate Performance: Full C++ stack from protocol core to GUI — no GC pauses, no runtime overhead, minimal memory and CPU footprint
Modern Codecs: H.264, VP8, VP9, AV1 — flexibly switch based on network and hardware conditions
WebRTC P2P Direct Connection: Prioritizes peer-to-peer connections for lowest latency; automatically falls back to TURN relay when traversal fails
Cross-Platform: Windows and macOS supported (Linux planned)
Modern UI: Fluent Design interface built with Qt 6 + QML, with light and dark themes

Comparison

| Feature | QuickDesk | RustDesk | BildDesk | ToDesk | TeamViewer | |---------|:---------:|:--------:|:--------:|:------:|:----------:| | AI Agent Support (MCP) | ✅ Built-in | ❌ | ❌ | ❌ | ❌ | | Open Source | ✅ MIT | ✅ AGPL-3.0 | ❌ | ❌ | ❌ | | Free for Commercial Use | ✅ | ❌ License required | ❌ | ❌ | ❌ | | Core Language | C++ | Rust + Dart | — | — | — | | Remote Protocol | Chromium Remoting (WebRTC) | Custom | Custom | Custom | Custom | | Protocol Maturity | ⭐⭐⭐⭐⭐ Google 10yr+ | ⭐⭐⭐ | ⭐⭐ | ⭐⭐⭐ | ⭐⭐⭐⭐⭐ | | P2P Direct | ✅ WebRTC ICE | ✅ TCP Hole Punching | ✅ | ✅ | ✅ | | Video Codecs | H.264/VP8/VP9/AV1 | VP8/VP9/AV1/H.264/H.265 | — | — | — | | Self-Hosted | ✅ Full solution | ✅ | ❌ | ❌ | ❌ | | GUI Framework | Qt 6 (C++) | Flutter (Dart) | — | — | — | | Memory Usage | ⭐⭐⭐⭐⭐ Very Low | ⭐⭐⭐ Medium | — | ⭐⭐⭐ | ⭐⭐⭐ | | CPU Usage | ⭐⭐⭐⭐⭐ Very Low | ⭐⭐⭐ Medium | — | ⭐⭐⭐ | ⭐⭐⭐ | | Windows | ✅ | ✅ | ✅ | ✅ | ✅ | | macOS | ✅ | ✅ | ✅ | ✅ | ✅ | | Linux | 🔜 | ✅ | ❌ | ✅ | ✅ | | iOS/Android | 🔜 | ✅ | ✅ | ✅ | ✅ |

Why Best-in-Class Performance?

Pure C++ Full Stack: The entire stack — from low-level protocol to GUI — is implemented in C++. No garbage collection pauses, no VM/runtime overhead. Compared to RustDesk's Dart/Flutter GUI layer or ToDesk's Electron-based approach, CPU and memory usage are significantly lower.
Chromium-Level Optimization: Core paths for video encoding/decoding, network transmission, and screen capture directly reuse Chromium's highly optimized C++ code, including SIMD instruction optimization and zero-copy rendering pipelines.
Shared Memory Video Transfer: QuickDesk uses shared memory to pass video frames (YUV I420) between processes, eliminating the serialization/deserialization and data copying overhead of traditional IPC.
GPU-Accelerated Rendering: YUV data is fed directly into the GPU rendering pipeline via Qt 6's QVideoSink, achieving zero-CPU-copy video rendering.

Features

AI Integration (MCP Server)

Built-in MCP Server — AI agents connect via standard protocol, works with Cursor, Claude Desktop, VS Code, and any MCP client
Dual transport: stdio (AI client launches process) or HTTP/SSE (QuickDesk manages MCP server, multiple AI clients connect)
Persistent MCP transport mode — remembers your stdio/HTTP choice across restarts
40+ MCP tools: screenshot, mouse click/drag/scroll, keyboard type/hotkey, clipboard, OCR text recognition, UI element detection, screen verification, and more
Host-side AI Agent with built-in skills (system info, file operations, shell execution) — run structured tools on the remote machine
Pluggable skills architecture — add custom skill directories, skills auto-load and sync to connected clients
AI Agent toggle in Settings — enable/disable the agent process with persistent configuration
MCP Resources: real-time device status, connection info, host details
9 MCP Prompts: operation guide, server health check, batch automation, system diagnosis, screen analysis, multi-device orchestration, SOP documentation
Real-time event streaming: connection state changes, clipboard updates, screen changes, performance stats
Event-driven tools: wait_for_event, wait_for_connection_state, wait_for_clipboard_change, wait_for_screen_change for reactive automation
Background automation mode: show_window=false for headless batch operations
Screenshot scaling: adjustable resolution for faster AI processing

Remote Control

High-definition, low-latency remote desktop display
Full keyboard and mouse mapping
Real-time remote cursor synchronization
Bidirectional clipboard sync
Adaptive frame rate and bitrate
Frame rate boost mode (Office / Gaming)

Connection Management

9-digit Device ID + temporary access code mechanism
Auto-refresh access code (configurable: 30 min to 24 hours, or never)
Multi-tab simultaneous connections to multiple remote devices
Connection history with quick reconnect
Real-time connection status monitoring

Performance Monitoring

Detailed latency breakdown panel (Capture → Encode → Network → Decode → Render)
Real-time frame rate, bitrate, and bandwidth statistics
Input round-trip time (RTT) monitoring
Encoding resolution and quality information

Personalization

Fluent Design style interface
Light and dark theme switching
i18n support (Chinese / English)
Video codec preference (H.264 / VP8 / VP9 / AV1)

Self-Hosted Deployment

Custom signaling server address
Custom STUN/TURN servers
Complete server deployment solution (Go signaling server + PostgreSQL + Redis + coturn)

Architecture

QuickDesk uses a modular multi-process architecture:

graph TD
    subgraph AI["AI Agent (Claude / GPT / Cursor)"]
        A1["MCP Protocol"]
    end

    subgraph MCP["quickdesk-mcp (Rust Bridge)"]
        M1["MCP stdio ↔ WebSocket"]
        M2["MCP HTTP/SSE ↔ WebSocket"]
    end

    subgraph GUI["QuickDesk GUI (Qt 6)"]
        direction LR
        WS["WebSocket API Server"]
        QML["QML/C++"]
        MainWin["Main Window"]
        RemoteWin["Remote Desktop Window"]
    end

    subgraph Host["quickdesk-host (Chromium Remoting)"]
        H1["Screen Capture"]
        H2["Video Encoding"]
        H3["Input Event Handling"]
        H4["WebRTC Connection"]
    end

    subgraph Client["quickdesk-client (Chromium Remoting)"]
        C1["Video Decoding"]
        C2["Input Event Injection"]
        C3["Shared Memory Video

QuickDesk

Install / Use

README