EXTRA - Native Computer Use

Name: EXTRA - Native Computer Use
Author: AIYantra

作者：AIYantra版本 0.2.5i★ –i在你的电脑上运行

注册名称io.github.AIYantra/extra

安装

概览

作者提供的说明（英文原文）

Flashless macOS & Windows native computer-use engine for AI vision, mouse & keyboard automation

PopNx MCP 的三大支柱

这三项永远不会合并成一个分数：意见、测量和人气是不同的东西。

用户评价i暂无评价评价功能将在后续阶段与账户功能一起开放
我们的基础检查i1 项中通过 1 项依据公开信息检查；尚未扫描代码
使用情况i尚未统计★ –

它能帮你做什么

我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。

连接方式: 在你的电脑上运行
传输方式: stdio
分类: 其他

安装

选择你使用的 AI 应用，然后复制设置。

安装之前

基础检查：1 项中通过 1 项
在你的电脑上运行，拥有你的用户账户的权限（文件、网络、应用）。
注册中心没有列出需要填写的密钥。

你需要

PyPI已安装 Python 的 uv（用于 uvx 命令）

你用的是哪个应用？

ClaudeClaude CodeCursorVS CodeChatGPT

在上方选择你的应用。我们会记住它，下次按钮会显示“安装到 …”。

打开 Claude Desktop → Settings → Developer → Edit Config，把下面的内容粘贴到 claude_desktop_config.json

{
  "mcpServers": {
    "extra": {
      "command": "uvx",
      "args": [
        "extra-desktop"
      ]
    }
  }
}

在终端中运行这条命令

claude mcp add extra -- uvx extra-desktop

在 Cursor 中打开

或者自己复制设置

粘贴到项目中的 .cursor/mcp.json，或 ~/.cursor/mcp.json

{
  "mcpServers": {
    "extra": {
      "command": "uvx",
      "args": [
        "extra-desktop"
      ]
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

在 VS Code 中打开

或者自己复制设置

粘贴到项目中的 .vscode/mcp.json

{
  "servers": {
    "extra": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "extra-desktop"
      ]
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

这个服务器在你的电脑上运行，而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。

你的选择通过 Cookie 保存在这台设备上，无需账户。

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同，请以 README 为准。

它能做什么

它在你的电脑上运行；安装后，工具列表会显示在你的 AI 应用中。

基础检查i

我们检查的是任何人都能看到的信息；我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日

✓ 通过
作者与仓库一致
注册名称已通过 GitHub 账户 AIYantra 验证，该账户拥有此仓库
… 尚未检查
仍在维护
我们还没有从 GitHub 读取这个仓库
… 尚未检查
有许可证
我们还没有从 GitHub 读取这个仓库
– 不适用
端点有响应
它在你的电脑上运行，没有互联网端点
– 不适用
使用 OAuth 登录
在你的电脑上运行，改用本地设置
… 尚未检查
工具描述中没有隐藏指令
它在你的电脑上运行；我们只能读取互联网端点的工具列表

与“其他”中的其他服务器比较i

星标数多于此分类中其他 32673 个服务器的 0%
发布到注册中心的时间比此分类中其他 32673 个服务器的 45% 更新
基础检查结果等于或优于此分类中其他 32673 个服务器的 100%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时，星标数才计入。

常见问题

EXTRA - Native Computer Use 安全吗？

它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前，请先看看这个服务器要求哪些权限，并使用只具备所需权限的密钥。

它需要哪些密钥或代码？

注册中心没有列出必需的密钥，不过有些服务器会在首次使用时要求你登录。

它在我的电脑上运行还是在云端？

它在你的电脑上运行。你的 AI 应用会在需要时启动它，数据不会经过作者的服务器（除非它自己调用在线服务）。

它还在维护吗？

GitHub 仓库最后更新于–，最新版本（0.2.5）于2个月前进入注册中心。

哪些应用可以使用它？

支持 MCP 的应用，例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Python (uv)。

有什么可以替代它？

同一分类中有：Spec Driven Development, Remote Desktop Commander, Systemdox

替代选择

同一分类中其他作者的服务器，按星标数排序。

如何看卡片：★ GitHub 星标i三大支柱：评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

Spec Driven Development

作者：formulahendry

MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.

在你的电脑上运行★ 440MIT14个月前更新

安装

Remote Desktop Commander

作者：desktop-commander

Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.

在云端运行★ 211前天更新

安装

Systemdox

作者：puglieseweb

Your architecture in your agents' context: checks, specs, ADRs and docs over MCP, at write time.

在你的电脑上运行今天更新

安装

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

Extra

The Open-Source Astra 6 for your Linux PC, Mac & Windows

Your AI can see, click, type, navigate, and get real work done on Linux, macOS & Windows.

Website • Quickstart • Linux Guide • macOS Guide • Architecture • License

What is Extra and Why Does It Exist?

Extra is an open-source Computer Use protocol and Model Context Protocol (MCP) server developed by yantraOS. It bridges LLMs (including ChatGPT, Claude, Cursor, and Gemini) directly to physical desktop operating systems across Linux, macOS, and Windows. Extra replaces fragile vision-only screenshot loops with native OS accessibility trees and sub-3ms GPU frame streaming, delivering pixel-perfect clicking, instant typing, and up to 225× lower token costs.

Today's AI assistants can write brilliant essays and code, but they are trapped inside a chat window. When you need them to click a button, open an app, organize your files, or input data into a spreadsheet, they can only give you text instructions.

Extra gives your AI hands and eyes on Linux, macOS & Windows.

It connects Claude, Google Antigravity, Cursor, AGY, or any autonomous agent directly to your desktop with high-speed screen vision, pixel-perfect clicking, and instant typing.

No clunky browser extensions.
No expensive cloud servers watching your screen.
100% open source, local, and private.

What Can an AI Assistant Do With Extra?

Once Extra is running, you can talk to your AI like a real human assistant sitting at your desk:

📊 Spreadsheets & Data: *"Open Excel or LibreOffice Calc, calculate total revenue from the invoice CSVs in my Downloads folder, and create a summary chart."*
🗂️ Desktop & File Cleanup: *"Clean up my messy desktop by moving screenshots into Pictures, PDFs into Documents, and deleting empty folders."*
🌐 Web Research & Data Entry: *"Open Chrome or Firefox, find the top 5 flights to Tokyo under $800, and copy their flight numbers and dates into a text note."*
⚙️ System Administration: *"Open Settings or terminal, check system updates, and verify service health."*
🎵 App Control: *"Launch Spotify or VLC, search for low-fi focus beats, and start playing."*

⚡ Quickstart (Under 10 Seconds)

You can set up Extra in seconds on Windows, macOS, or Linux:

Option A: 1-Click Cursor Install (Zero Terminal)

If you use Cursor IDE, install Extra with a single click via the native deeplink:

👉 [Add Extra to Cursor (1-Click Install)](cursor://anysphere.cursor-deeplink/mcp/install?name=extra&config=eyJjb21tYW5kIjoidXZ4IiwiYXJncyI6WyJleHRyYS1kZXNrdG9wIl19)

Cursor will prompt: *"Install MCP Server 'extra' with command 'uvx extra-desktop'?"* — click Install and you're ready!

Option B: Automated One-Liner (Configures All Clients Simultaneously)

The one-liner downloads the lightweight core engine (< 6.3 MB) and automatically configures Cursor, Claude Desktop, Windsurf, Claude Code, and Antigravity with zero JSON editing:

🪟 Windows PowerShell:

  irm https://extra.yantraos.com/install.ps1 | iex

🍏 macOS Terminal:

  curl -sSL https://extra.yantraos.com/install.sh | bash

Or ask your AI on macOS directly:

  Setup Extra on my Mac: In Terminal run 'curl -sSL https://extra.yantraos.com/install.sh | bash', then read and configure ~/.extra/app/STARTER_PROMPT_MACOS.md so we are ready to use Extra.

🐧 Linux Terminal (Arch / yantraOS, Ubuntu, Debian, Fedora):

  curl -fsSL https://extra.yantraos.com/install.sh | bash

The automated installer will:

Verify OS architecture (Windows 10/11 64-bit, macOS 12.3+, or Linux Wayland/X11).
Provision isolated Python runtime (~/.extra/venv) in under 3 seconds.
Auto-detect and wire Cursor, Claude Desktop, Windsurf, Claude Code CLI, and Antigravity (agy).
Run the health diagnostic doctor and confirm readiness.

Option C: Instant Zero-Install Runner (`uvx` / `npx`)

Run Extra instantly without installing or configuring local venvs:

# Via Astral uv (sub-second cold start):

uvx extra-desktop run

Or via Node npx:

npx -y @yantraos/extra-desktop

🎯 Your First 3 Missions

Once Extra is connected, paste any of these 3 verified starter missions into your AI to verify its hands and eyes:

🪄 The Desktop Magic Trick (Window Snap + Notepad Haiku):

*"Snap my active window to the left, open Notepad on the right, and write a haiku about autonomous desktop agents."*

🧮 The Fast Calculator (Sub-3ms Math):

*"Open Calculator, calculate 245.12 * 3.14159, and tell me the result."*

📁 The Safe File Organizer (Desktop Triage):

*"Inspect my Desktop, find all screenshot files, and organize them into a clean Screenshots folder."*

Option D: Developer Git Clone

# Linux / macOS:

git clone https://github.com/AIYantra/extra.git ~/.extra/app cd ~/.extra/app && ./install.sh

Windows:

git clone https://github.com/AIYantra/extra.git "$HOME\.extra\app" cd "$HOME\.extra\app" ; .\install.ps1

Connecting to Any MCP-Compatible AI

Extra works out of the box with any agent supporting the Model Context Protocol (MCP). If you use Cursor, Windsurf, or custom agents, add this snippet to your MCP config:

"mcpServers": { "extra": { "command": "npx", "args": ["-y", "@yantraos/extra-desktop"] } } }

Or using native Python:

"mcpServers": { "extra": { "command": "python", "args": ["-m", "extra.mcp.server"] } } }

How Does Extra Work Under the Hood?

Extra operates via a sovereign Dual-Plane Architecture that decouples visual perception from hardware actuation. For perception, Extra combines sub-3ms GPU frame capture (DirectX DXGI, Metal, Spectacle) with OS accessibility hierarchy inspection (COM UI Automation, AT-SPI2, Quartz). For actuation, Extra dispatches lossless kernel and Win32 hardware events with zero coordinate drift, guaranteeing deterministic control across all resolutions and display scaling factors.

Extra is engineered from the ground up for speed, reliability, and token efficiency:

👁️ Ultra-Fast Screen Capture (< 3ms): Uses native DirectX Desktop Duplication (DXGI) to take crystal-clear desktop frames in under 3 milliseconds—without lagging your PC or blurring text.
🎯 Pixel-Perfect Clicking: Rather than guessing coordinates from fuzzy screenshots, Extra queries the native Windows accessibility tree (UI Automation) to click the exact button, menu, or text field with 100% mathematical accuracy.
⚡ Instant Typing: Types 500 characters in under 5 milliseconds via native Win32 Unicode injection—with zero dropped letters and full support for emojis and international languages.
🛑 Infinite Loop Stall Protection: If an app freezes or a click produces no visual result, Extra immediately catches it and stops safely instead of burning your tokens in an endless loop.
🔒 100% Local & Private: Extra runs completely on your machine. Zero screenshots, keystrokes, or telemetry are ever sent to any cloud server.

📊 Performance Comparison

Metric / Capability	Legacy Scripts (`PyAutoGUI`)	Cloud Vision Models	Extra Engine
Screen Grab Speed	150 – 300 ms	80 – 150 ms	1.8 – 3.2 ms (DirectX DXGI)
Typing Speed (100 chars)	2.5 – 5.0 seconds	1.0 – 2.0 seconds	< 0.005 seconds (Instant Win32)
Click Accuracy	~60% (fails on display scaling)	~82% (vision guess)	99.4% (Native Windows UIA)
Multi-Monitor DPI Support	Broken on 125%/150% scales	Requires manual adjustment	Automatic (PerMonitorV2)
AI Token Cost	High (full screenshot every step)	High (full vision payload)	70% Lower (Smart element tree)
Stall Prevention	None (gets stuck forever)	Basic timeout	Smart visual delta detection

🛠️ Included Tools (MCP Suite — 20 Native Tools)

When connected to Extra, your AI assistant receives full sovereign desktop control across 20 native tools:

Tool Name	What It Does
`extra_launch`	Opens any app, utility, or URL directly (`calc`, `notepad`, `settings`, `edge`, `chrome`)
`extra_focus_window`	Brings any window to the front via native thread input attachment
`extra_click`	Clicks with sub-pixel DPI accuracy, human-like Bezier curves, and semantic visual grounding (`target="..."`)
`extra_stroke`	Draws fluid, continuous brush splines across waypoints with pressure and Catmull-Rom smoothing
`extra_type`	Injects text instantly with zero lag, full emoji support, and Win32 Unicode packets
`extra_hotkey`	Sends keyboard shortcuts (`Ctrl+C`, `Win+E`, `Alt+Tab`, `Enter`)
`extra_batch_actions`	Executes atomic lists of hardware & SOUL reflex actions (`eval`, `assert`, `wait_for_state`) in sub-milliseconds
`extra_snap_layout`	Programmatically docks and arranges windows in side-by-side or split layouts in < 15ms
`extra_fs_batch`	High-speed batch filesystem operations compliant with Windows Defender Controlled Folder Access
`extra_inspect_ui`	Scans accessible UI trees and interactive element nodes in real time
`extra_click_element`	Clicks UI elements deterministically by accessible ID or semantic target
`extra_scroll`	Smoothly scrolls wheels up, down, left, or right
`extra_drag`	Drags and drops files, windows, or sliders between coordinates
`extra_screenshot`	Captures high-res desktop frames in < 25ms with optional Set-of-Mark visual badges
`extra_browser`	Directly extracts DOM content in Edge/Chrome or triggers synthesized API fast-paths
`extra_task_start`	Activates ambient screen edge pulse and cursor tracking halo
`extra_task_complete`	Signals task completion: flashes emerald border, plays acoustic chime, and analyzes trajectory
`extra_indicate_status`	Direct programmatic control over active, complete, and idle desktop indicators
`extra_recall_memory`	Recalls past task workflows, artifacts, and known quirks via KùzuDB + FastEmbed in < 2ms
`extra_scout_app`	Discovers UI frameworks (Electron/Win32/Viewport), universal hotkeys, and generates SKILL playbooks
`extra_evolve_skill`	Crystallizes newly verified zero-stall fast paths into permanent skill playbooks

🏛️ Repository Structure

extra/

├── assets/ # Brand logos and banners ├── bin/

只显示了开头部分；其余内容请在仓库中阅读