Remote Desktop Commander
Build, explore, automate on your computer — from anywhere.
Reach your computer's filesystem and terminal from any AI app that supports remote MCP connectors. Remote Desktop Commander is a hosted MCP server that relays a secure connection to your real machine, giving your AI assistant the same full file and shell access as the local server — without you needing to be at the keyboard.
This repository hosts the public manifests, documentation and issue tracker for the hosted service. The service implementation is not open source. For the open-source local (stdio) server, see DesktopCommanderMCP.
What you can do
- Build from scratch — create features and applications with simple commands
- Manage development environments — set up servers, configure systems, handle processes
- Explore existing codebases — navigate and understand complex repositories
- Edit documents in place — work on Word, Excel, and PDF files where they live on disk
- Manage context and documentation — keep track of project details and technical specifications
- Control multiple machines — pair several computers and address them by name in one conversation
Quick start
1. Start the device agent
On the computer you want to control:
npx @wonderwhy-er/desktop-commander@latest remote
Your browser opens to a verification page showing a pairing code (also printed in the terminal — a standard OAuth device-flow check). Sign in, confirm the codes match, and the terminal reports "Device ready". Keep the agent running: close the terminal, no access.
2. Connect your AI client
Each client walks you through OAuth sign-in on first connect.
claude.ai / Claude Desktop — Settings → Connectors → Add custom connector, then enter:
https://mcp.desktopcommander.app/mcp
Claude Code:
claude mcp add --transport http remote-desktop-commander https://mcp.desktopcommander.app/mcp
(If you work inside a checkout of this repo, Claude Code also picks up the included .mcp.json automatically.)
ChatGPT — Settings → Apps & Connectors → Advanced settings → enable Developer mode (requires a plan that includes it), then Create connector and enter:
https://mcp.desktopcommander.app/mcp
Cursor — one click: Add to Cursor. Or add to ~/.cursor/mcp.json manually:
{
"mcpServers": {
"remote-desktop-commander": {
"url": "https://mcp.desktopcommander.app/mcp"
}
}
}
VS Code — install the agent plugin (bundles this server):
copilot plugin install desktop-commander/remote-desktop-commander
Or add the server directly to .vscode/mcp.json in your workspace (or your user-level mcp.json):
{
"servers": {
"remote-desktop-commander": {
"type": "http",
"url": "https://mcp.desktopcommander.app/mcp"
}
}
}
Gemini CLI:
gemini extensions install https://github.com/desktop-commander/remote-desktop-commander
3. Try it
These work against any machine running the agent, no setup needed:
"List the contents of my home directory and tell me what's there."
"Create a folder called ai-test in my Documents folder. Inside it, write a short Python script that prints the date, and run it."
"Start a Python REPL on my machine, import pandas, and tell me which version is installed. Keep the session alive."
"Search my home directory for files modified in the last 24 hours and group them by folder."
Full walkthrough with troubleshooting: docs/SETUP.md
Requirements
- Node.js 18 or newer (for
npx) on each computer you want to control
- One terminal running the device agent per machine
- Works on macOS, Windows, and Linux
| Category | Tools |
|---|
| Read & explore | Read File, Read Multiple Files, List Directory, Get File Info |
| Search | Start Search, Get More Search Results, Stop Search, List Active Searches |
| Write & edit | Write File, Edit File Block, Move or Rename File, Create Directory, Write or Modify PDF |
| Terminal & processes | Start Process, Send Input to Process, Read Process Output, List Active Sessions, List System Processes, Terminate Session, Kill System Process |
| Devices & account | List Paired Devices, Current User Info, Ping Device, Shutdown Device Agent |
| Configuration & help | Get Configuration, Set Configuration Value, Get Prompt Library, Get Recent Tool Activity, Get Usage Statistics, Submit Feedback |
How it works
sequenceDiagram
participant AI as AI app (web, mobile, desktop)
participant S as mcp.desktopcommander.app
participant D as Your computer (device agent)
AI->>S: MCP tool call (OAuth 2.0 bearer)
S->>D: relayed to your paired device
D->>D: executes locally
D->>S: result
S->>AI: MCP response
- Authentication is standard OAuth 2.0 with PKCE; devices pair via the OAuth device authorization flow (the short-code approval you see at setup).
- The server only accesses data explicitly requested through tool calls; transmission is encrypted in transit (HTTPS/TLS).
- A machine is reachable only while its device agent is running — stop it with
Ctrl+C and access ends. Devices can also be reviewed and revoked at any time from the dashboard.
- Tools execute with your user account's permissions on the paired machine, so treat your connected AI accounts like credentials and protect them with MFA. Read the full trust model in SECURITY.md.
Support & legal