AI Agent Hub
Back to plugins
🖥️

dsh-mac-control

Client Updated 2026.08.22

Run the following command in DeepSeek Harness:

dsh plugin install 18126295767-cell/dsh-mac-control

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install 18126295767-cell/dsh-mac-control in your DeepSeek Harness terminal to install the plugin, source available at https://github.com/18126295767-cell/dsh-mac-control ; after installation, restart the web profile to enable the mac_browser and mac_desktop tools.

About this plugin

If you run DeepSeek Harness locally, you already appreciate how smoothly it can chat, write code, and explain concepts. What it cannot do is put its hands on your Mac. Ask it to open a link, inspect a window, or grab a screenshot, and it can only narrate the steps. dsh-mac-control closes that gap.

It is a native macOS plugin with exactly two tools. mac_browser handles Safari and Google Chrome tabs: inspect, open, and switch. mac_desktop reaches into local apps and windows to read accessibility controls, type text, click, press key combinations, and capture the screen. Under the hood it starts no HTTP server, proxies no browser traffic, and stores no credentials. Every request is validated and then handed to macOS-native osascript (JXA/System Events), open, and screencapture, while the system continues to govern Automation, Accessibility, and Screen Recording permissions.

Safety is baked in by default. Only http and https URLs are accepted. Browser and desktop targets can be restricted via an allowlist. Text length and pointer coordinates are validated. The tool descriptions explicitly forbid sending passwords, API keys, cookies, or recovery codes, so the constraint lives in the model instructions rather than relying on user discipline.

If you are a macOS user who runs DeepSeek Harness in a local terminal and would like it to be more than a talking head, to actually switch a tab, read a UI, and snap a screenshot, this plugin is built for you. It is lightweight, service-free, credential-free, and fully compatible with the existing system permission model.

Use Cases

  • Ask the AI to read the current browser tab and summarize the page content
  • Have the AI switch to a target app window, read its controls, and type a response
  • Let the AI perform a sequence of desktop actions and then capture the screen as a log

Best For

  • macOS developers running DeepSeek Harness locally
  • Productivity users who want the AI to act on the desktop instead of just listing steps
  • Users who want to give the AI system-level capabilities without adding extra services or proxy layers