AgentToolDrop

36 of 43

agent-desktop

Gives an agent a CLI that reads and operates macOS apps through their accessibility trees

Install with npm

npm install -g agent-desktop        # downloads prebuilt binary automatically

Copied from the project's docs at github.com/lahfir/agent-desktop

Why you'd want it

agent-desktop is a Rust CLI that snapshots an app's accessibility tree into compact element references, which the agent then uses to click, type, scroll and manage windows. A skeleton mode returns a shallow overview of dense apps such as Slack and lets the agent drill into one region, and commands return structured JSON with error codes and recovery hints. Sessions record traces that export to a single HTML viewer, and launch --cdp opens a DevTools port on Chromium-based apps so Playwright or Puppeteer can drive the web contents.

Best for

  • Agent control of native Mac apps
  • Compact UI snapshots of dense apps
  • Calling from Python, Swift or Node through its C library

Use it when

You want a shell-driven agent to operate Mac apps by element reference instead of guessing coordinates from screenshots.

Skip it if

You are on Windows or Linux, which the README lists as planned. It needs macOS Accessibility permission, plus Screen Recording for screenshots, and can click, type and use the clipboard in any app it is pointed at.

Related tools

Same category or shared tags.

  1. UI-TARS Desktop

    Browser agent Standard

    Controls your computer and browser from screenshots, with mouse and keyboard actions chosen by a vision model

    • 39.3k stars
    • 62 stars/day since creation
    • pushed
    • v0.3.1 on
    • Apache-2.0

    Download from GitHub releases or install with Homebrew

    brew install --cask ui-tars
    Use it when, skip it if, stats for UI-TARS Desktop

    Why you'd want it

    The repo ships two ByteDance projects. UI-TARS Desktop is a desktop app with a native GUI agent driven by the UI-TARS and Seed-1.5-VL/1.6 vision-language models: it takes screenshots, recognizes what is on screen and controls the mouse and keyboard to follow natural-language instructions, with local computer and browser operators. Agent TARS is a separate CLI and Web UI agent that controls browsers through visual grounding, the DOM or a hybrid of both, and can mount MCP servers to reach other tools.

    Best for

    • Experimenting with vision-based computer use
    • Automating desktop apps that have no API
    • Running an open GUI agent model you deploy yourself

    Use it when

    You want an agent that operates desktop apps or a browser by looking at the screen, and you can point it at a UI-TARS model endpoint, for example one deployed on Hugging Face.

    Skip it if

    You use more than one monitor (the quick start says only single-monitor setups are supported) or want a library to embed in your own code. The free remote operator service was discontinued on August 20, 2025.

    Stats

    Stars
    39,279
    Last push
    Release
    v0.3.1,
    License
    Apache-2.0
    Pricing
    Open source
    Works with
    macOS, Windows
    Stars per day, since repo creation
    62
    Tier
    Standard, not ranked

    Full stat block for UI-TARS Desktop

  2. Browser Use

    Browser agent Standard

    Lets an LLM drive a real browser to complete multi-step web tasks from Python code or a CLI

    • 118k stars
    • 165 stars/day since creation
    • pushed
    • 0.13.11 on
    • MIT

    Install with uv

    uv add browser-use
    Use it when, skip it if, stats for Browser Use

    Why you'd want it

    Browser Use is an open-source browser agent: you give it a task in plain language and a model, and it opens a browser, works through the site and returns a result. It works with OpenAI, Anthropic and Google models or local models through Ollama, accepts custom tools registered as Python functions, and can drive a local Chrome profile, a remote browser over CDP or a Browser Use Cloud browser. A CLI gives existing agents such as Claude Code, Codex and Cursor browser access, and an integration implements Claude's browser toolset with Browser Use as the driver.

    Best for

    • Python apps that automate tasks on real websites
    • Giving a coding agent browser access
    • Projects that may move to hosted browsers later

    Use it when

    You want an agent that completes tasks on live websites, such as forms, bookings and lookups, from your own Python code with your choice of model.

    Skip it if

    You only need fixed, scripted browser steps with no model in the loop. The library needs Python 3.11 or later, and stealth browsers, proxies and the BU2 model are paid Browser Use Cloud services.

    Stats

    Stars
    118,013
    Last push
    Release
    0.13.11,
    License
    MIT
    Pricing
    Open source
    Works with
    Claude Code, Codex, Cursor
    Stars per day, since repo creation
    165
    Tier
    Standard, not ranked

    Full stat block for Browser Use

  3. Rank 9

    Browser Harness

    Browser agent

    Lets Claude Code, Codex or Cursor drive your own Chrome over CDP, logged in as you

    • score 53
    • 18.4k stars
    • 102 stars/day since creation
    • pushed
    • v0.1.13 on
    • MIT

    Install with uv

    uv tool install --python 3.12 --upgrade --force browser-harness
    Use it when, skip it if, stats for Browser Harness

    Why you'd want it

    Browser Harness connects a coding agent to your running Chrome through the DevTools Protocol, so it can open pages, click, type, read and download with your existing logins. It installs as a CLI with a skill that teaches the browser workflow, and the agent writes missing helpers into its local workspace as it works while the core package stays protected. A browser-harness-mcp command exposes the same helpers as MCP tools for any MCP client.

    Best for

    • Logged-in web tasks from a coding agent
    • Claude Code, Codex and Cursor
    • MCP clients that need a real browser

    Use it when

    You want your coding agent to work in the browser where you are already signed in, instead of a fresh headless profile.

    Skip it if

    You do not want an agent acting in your real, logged-in browser profile: setup turns on Chrome remote debugging, and the agent acts as you on the sites you use. Optional local recordings save screenshots and action traces that may include sensitive page content.

    Stats

    Stars
    18,433
    Last push
    Release
    v0.1.13,
    License
    MIT
    Pricing
    Open source
    Works with
    Claude Code, Codex, Cursor, Any MCP client
    Stars per day, since repo creation
    102
    New and rising score
    53 of 100

    Full stat block for Browser Harness

For agents

The same entry, as files an agent can read without scraping this page. 

# read one entry
curl -s https://agenttooldrop.com/tools/agent-desktop.md

# list the MCP servers in the catalog
curl -s https://agenttooldrop.com/tools.json \
  | jq -r '.tools[] | select(.category=="mcp-server") | .slug'