AI Agent Hub
Back to plugins
🖥️

auto-mouse

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install Fish121380/auto-mouse

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

Run dsh plugin install Fish121380/auto-mouse in DeepSeek Harness to fetch the plugin from https://github.com/Fish121380/auto-mouse, register the MCP endpoint, and confirm port 49153 to start using it.

About this plugin

Large language models can reason about almost anything, yet they still cannot simply see your desktop. auto-mouse (Little Mouse Pointer) closes that gap: it is a lightweight Windows desktop context picker built specifically for AI agents. You drag a box around a window, a control, or any screen region, and the tool harvests structured context through three local channels—UI Automation tree, a high-resolution screenshot, and on-device OCR—then hands the result to OpenAI Codex, DeepSeek Harness, or any MCP-compatible client. Everything runs locally; no pixel ever leaves your machine.

The workflow revolves around three MCP tools. ui_context_pick initiates an interactive selection and returns the window title, control hierarchy, screenshot, and recognized text. ui_context_current lets you query the cached context snapshot at any time. ui_context_clear wipes the cache once the task is done. The distributable is a self-contained Windows x64 binary—no .NET runtime to install—so wiring it into an existing MCP configuration is a one-line change.

It is built for developers who repeatedly need an AI to understand the state of a GUI: reading an error dialog, mapping a form's field structure, or analyzing a layout snippet. If you want that capability without sending screen data to a third-party OCR service, this local, plug-and-play picker slots neatly into your AI toolchain.

Use Cases

  • Box-select an error dialog so the AI reads control text to pinpoint issues
  • Capture a form region and let local OCR extract field structure for code generation
  • Pick a window layout and ask the model to analyze hierarchy and suggest improvements

Best For

  • Developers building AI coding assistants with DeepSeek Harness or Codex
  • Users who need AI to interpret desktop GUI state without sending screen data to the cloud
  • Engineers constructing desktop automation workflows on Windows