mac-computer-use
A macOS desktop control skill: the agent sees windows, reads elements, clicks, types and drives Chromium over CDP
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Runs local commands and controls apps: it clicks, types and changes program state
- Requires the host's accessibility and screen recording permissions
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/mac-computer-useDetects the agents on your machine, checks the risk and pins the version.
Other ways to install
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a claude-code -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a cursor -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a github-copilot -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a codex -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a gemini-cli -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a cline -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a roo -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
A fork of Roo Code, same .roo folders.
Run in a terminal in the project folder
npx skills add to3akarin/mac-computer-use --skill mac-computer-use -a opencode -yThe skills tool installs the current version from the repository. Add the -g flag to use the skill in every project.
Clone the repo, build the binary with bash scripts/build.sh and check the environment via .build/release/mac-computer-use doctor. In system settings grant the terminal accessibility and screen recording permissions.
Other ways from the author
git clone https://github.com/To3akaRin/mac-computer-use.git
cd mac-computer-use
bash scripts/build.shBuilds the native binary. Then check the environment via doctor.
This is third-party code. Review the repository files before installing.
What it does
The skill gives the agent tools to work with real macOS applications. The agent lists windows and processes, reads the accessibility tree, takes screenshots and performs clicks, typing and shortcuts. For embedded Chromium pages there is control over the CDP protocol: reading structure, inserting text and running read-only assertions. The loop is built on observation and verification: inspect the target, pick a channel, run one action, then read the result and confirm it. Write commands support a dry run, and stale window references are rejected.
Who it is for. For engineers and testers who need macOS desktop app automation with result verification.
Good fit when
- You need to automate a desktop app that has no ready API
- You need UI acceptance with screenshots and state reads
- You need to run an action in an app and confirm it applied
Not a fit when
- A plain web task where a browser tool is enough
- You have no Mac or no accessibility and screen recording permissions
- The environment is only a remote Linux sandbox with no access to a host Mac
Example request
Find this app's window, type text into the right field and read it back to verifyLimitations
It runs only on macOS 14 and newer, needs Swift 6 and accessibility and screen recording permissions; CDP also needs Node.js 22.4+. The first native call compiles the binary, which is an expected wait. A test app and an isolated browser are verified; not every third-party program is guaranteed to work.
How to disable. Remove the mac-computer-use skill folder from your agent's skills directory. Optionally revoke the host app's accessibility and screen recording permissions in system settings.
Security check
- Runs local commands and controls apps: it clicks, types and changes program state
- Requires the host's accessibility and screen recording permissions
README in short
The README presents mac-computer-use as a skill under the open Agent Skills format that gives an agent direct control of a Mac. It lists observing windows and processes, reading and changing accessibility elements, native clicks and typing, plus embedded Chromium control over CDP. It stresses execution boundaries: dry-run write commands, explicit coordinate units, orderly foreground use and returning an unknown status when a result cannot be confirmed. There is a shared entry via scripts/run.sh for different clients and a test suite. MIT licensed.
SKILL.md
--- name: mac-computer-use description: 操作 macOS 桌面应用,探测窗口和自动化接口、截图、读取或修改辅助功能元素、执行鼠标键盘动作,以及通过 CDP 操作内嵌 Chromium 页面。适用于桌面应用自动化与界面验收;普通网页任务优先使用已有浏览器工具。 license: MIT compatibility: 需要 macOS 14+、Swift 6+、可写技能构建目录,以及读取文件和执行本机命令的能力;CDP 需要 Node.js 22.4+。桌面控制需要宿主的辅助功能和屏幕录制权限。 metadata: author: To3akaRin version: "0.2.0" --- # Mac Computer Use 本技能使用本目录中的独立工具,不需要专属 MCP 插件。所有相对路径均以本文件所在目录为基准;Agent 的当前目录可能是其他项目,执行时先解析技能根目录并使用绝对路径。 ## 通用入口 本目录遵循开放的 Agent Skills 格式。使用 runtime 提供的文件读取、本机命令执行和图片查看能力;agents/openai.yaml 是可选界面元数据,不参与核心运行。 ```bash sh "$SKILL_DIR/scripts/run.sh" native doctor sh "$SKILL_DIR/scripts/run.sh" native windows sh "$SKILL_DIR/scripts/run.sh" cdp targets --endpoint http://127.0.0.1:9222 ``` ## 开始前 1. 通过统一入口运行 native doctor 检查平台与权限。 2. 明确目标应用、窗口与用户要得到的结果。 3. 先 probe 和观察。优先应用已有 CLI、脚本字典或接口,其次 AX,再考虑鼠标键盘。 ## 操作循环 观察目标 选定元素或坐标 写操作预演 执行一次 读回目标状态 判断完成或重新观察。
FAQ
Do I need a separate MCP plugin?
No. The skill uses its own bundled tools and works in any runtime that reads SKILL.md and runs local commands.
What does the dry run do?
Write commands with the dry-run flag check parameters and preconditions but do not type, click or restart the app.
Related
An MCP server with n8n node and template knowledge: the agent picks nodes, validates configs and, with API access, creates workflows in your n8n
Zapier's official MCP plugin: the agent gets actions across thousands of apps through your Zapier account
Awesome Claude Skills by Composio
Awesome Claude Skills
A curated list of Claude skills and plugins, plus Composio's own automation skills for 78 SaaS apps
A plugin and CLI catalog for agents: generates command-line interfaces for GUI apps like GIMP and Blender and installs ready ones via CLI-Hub