dexbox

name	dexbox
description	Control Windows VMs and RDP desktops via dexbox
metadata	{"openclaw":{"emoji":"","requires":{"bins":["dexbox"]},"install":[{"id":"dexbox-curl","kind":"shell","command":"curl -sSfL https://raw.githubusercontent.com/getnenai/dexbox/main/install.sh | sh","bins":["dexbox"],"label":"Install dexbox via curl"}]}}

Dexbox Desktop Control

Dexbox lets you control Windows VMs and RDP desktops programmatically. You have 12 tools available split into two categories.

Prerequisites

Before using dexbox tools, the dexbox server must be running with at least one desktop connected:

dexbox start          # Start the dexbox server
dexbox up desktop-1   # Connect a desktop

If dexbox is not installed, install it with:

curl -sSfL https://raw.githubusercontent.com/getnenai/dexbox/main/install.sh | sh

Or see the README for other install methods.

Workflow

1. Discover desktops: Call list_desktops to see available desktops and their states
2. Start if needed: Call start_desktop to bring a desktop online
3. Screenshot first: Always call screenshot before interacting to see the current screen
4. Interact: Use click, type_text, key_press, scroll to control the GUI, or bash for PowerShell commands
5. Verify: Take another screenshot after actions to confirm the result
6. Prefer PowerShell: Use bash for tasks that can be done programmatically — it's faster and more reliable than GUI interaction

Desktop routing

All action tools accept an optional desktop parameter:

• One desktop connected: Omit the parameter — the server auto-resolves
• Multiple desktops: Pass the name from list_desktops to target a specific desktop

Tool reference

Lifecycle tools

Tool	Description
list_desktops	List all desktops with state and connection info
create_desktop	Register a new RDP desktop connection
destroy_desktop	Delete a VM or unregister an RDP connection
start_desktop	Boot a VM or connect an RDP session
stop_desktop	Shut down the VM guest OS or disconnect RDP
status_desktop	Get status of a single desktop

Action tools

Tool	Description
screenshot	Take a screenshot (returns PNG image)
click	Click at x,y coordinates (left/right/middle/double)
type_text	Type a text string (click the input field first)
key_press	Press a key or combo (e.g. enter, ctrl+c, alt+tab)
scroll	Scroll at x,y with direction (up/down) and amount
bash	Run a PowerShell command in the guest VM

Architecture

All tool calls go to the local dexbox server (default http://localhost:8600). Action tools use model=claude-mcp as an internal wire format selector to route through the server's Anthropic adapter for request/response parsing. This is a local format choice, not an external model call. All data (screenshots, command output) stays between this plugin and the dexbox server on your machine.

Tips

• This is a Windows VM — use Windows-style paths (e.g. C:\Users\dexbox\Desktop\file.txt)
• The bash tool runs PowerShell, not Unix bash
• When clicking, target the center of UI elements from the screenshot
• If an action doesn't produce the expected result, take a screenshot and try an alternative approach
• For typing, always click the target input field first