Skip to main content
Glama

Sabir lets AI agents — Claude, Codex, Gemini CLI and any MCP client — use Windows apps, including apps that have no API.

Instead of taking screenshots and guessing where to click, Sabir reads an app's interface through Windows UI Automation, saves every function it finds (buttons, menus, text fields, switches) as a reusable tool, and serves those tools to your agent over MCP. Every agent asks you once per app (Always allow / Allow once / Deny), and every action lands in the activity log.

Why not just computer use?

Computer use

Sabir

How it finds a button

Screenshot → look → guess pixels

Reads the UI tree, clicks the control by its automation ID

Screenshots per action

1 (~1,365 image tokens at 1280×800)

0

Your mouse and screen

Taken over

Stays yours — Sabir works in the background

Next time

Figures the app out again

Learned once, reused by every agent

Measured on a Windows 10 PC: learning Calculator (37 functions) took 235 ms; typing 12 + 30 = and reading the result took 84 ms; writing a 7-line list into Notepad took 63 ms. (That is Sabir's execution time — your agent still needs its own model calls to decide what to do.)

Related MCP server: omni-computer-use

Install

  1. Download Sabir-win-x64.zip from the latest release and unzip it anywhere.

  2. Start Sabir.exe. It lives in the tray (click the gear icon). The build is not code-signed yet, so Windows SmartScreen may warn you — choose More info → Run anyway.

  3. In the setup, click Connect for your agent. Sabir writes the MCP entry for you:

    • Claude Code — user-scoped HTTP server (claude mcp add-json), Claude Desktop — stdio bridge

    • Codex — ~/.codex/config.toml

    • Gemini CLI — ~/.gemini/settings.json

    • Other agents — Copy config gives you a ready MCP snippet

  4. Restart your agent once and ask it, e.g. "Open Notepad and write a shopping list."

Requires Windows 10 or 11. Nothing else to install — the download is self-contained.

MCP tools

sabir_list_apps · sabir_open_app · sabir_learn_app · sabir_list_tools · sabir_use_tool · sabir_read_app · sabir_act · sabir_press_keys

…plus one tool per learned function, e.g. notepad__file_save, notepad__set_text_editor, calculator__equals. New tools are announced to connected agents right away (tools/list_changed).

Security

  • The MCP server listens on 127.0.0.1 only and requires a random bearer token (stored in %APPDATA%\Sabir\config.json). Browser origins other than localhost are rejected.

  • Agents need your permission per app. Allow once lasts until the agent disconnects; Deny blocks the app for that session.

  • Pause stops all agent actions immediately (tray → Pause).

  • Everything is logged in %APPDATA%\Sabir\activity.jsonl. Before Sabir edits an agent config file, it keeps a .sabir-backup copy.

Build from source

npm install
npm run build:uia   # C# UI Automation helper (needs the .NET 9 SDK)
npm start           # run in development
npm run dist        # standalone build: dist/Sabir-win32-x64/Sabir.exe

Part

Where

Electron main process: tray, windows, permission dialog

main.js

Screens (tray, main window, permission, setup)

ui/

UI Automation helper (C#)

uia/Program.cs

Learning, tools, permissions, log

core/engine.js, core/store.js

MCP server (Streamable HTTP)

core/mcp.js

stdio bridge for stdio-only MCP clients

bridge/sabir-mcp.js

Agent config writer

core/connect.js

Credits

Fonts: Cinzel, Barlow Condensed, Inter — SIL Open Font License (see ui/fonts). Built with Electron and .NET.

License

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI coding agents to automate Windows desktop applications through semantic UI Automation instead of brittle coordinate clicks, with tools for discovering windows, finding controls by stable identifiers, and verifying actions.
      41 PyPI
      6
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables Windows desktop automation via MCP, allowing AI agents to control mouse, keyboard, and screen capture with the same interface as Anthropic's computer-use tool.
      64 PyPI
      8
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      MCP server exposing Windows desktop UI Automation as tools for AI agents, enabling automated interaction with Windows applications via Appium.
      90
      38 npm
      1
      Apache 2.0
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables AI agents to inspect and operate Windows applications through structured UI Automation maps, humanized mouse and keyboard input, and on-demand screenshots or OCR via MCP. It supports locating elements by role, name, or query; clicking, typing, scrolling, window management, event watching, and chained multi-step actions without relying on image analysis.
      MIT