# Deskwright

Hands an agent a real GNOME Wayland desktop: launching apps, clicking, typing, window handling and screen recording. A hidden virtual monitor keeps your focus intact.

- MCP server · `type=mcp` · Claude Code, Codex, Cursor, GitHub Copilot, Windsurf (Devin Desktop), Cline, Zed, Gemini CLI · ★ 11
- Author: tristanmuzzu
- License: Apache-2.0 · policy=link-only · description only; the file stays with the author
- Source: https://github.com/tristanmuzzu/deskwright
- Updated: 2026-10-07
- HTML: https://affpapa.org/en/nevibecode/mcp/deskwright

## What it does

A coding agent can edit files and run commands, yet it cannot open GIMP or press a button in a settings dialog. Deskwright fills that gap on GNOME. An MCP server plus a shell extension pass the live desktop to the agent: apps get launched, widgets get read, clicks and keystrokes go through, windows move, and OCR or screen recording is available.

Clicks do not aim at coordinates. The server triggers a widget's own accessibility action through AT-SPI, so a click cannot miss and survives the window moving. With `DESKWRIGHT_SESSION=headless` the whole session lands on a virtual monitor that is not on any of your screens, which is why the agent never grabs your mouse.

GNOME Shell on Wayland is required. Install with `pipx` and the `--system-site-packages` flag: PyGObject ships no wheels on PyPI, so an isolated environment cannot import it. Skip the flag and the server starts fine, then dies on the first click with `input_backend_failed`. After `deskwright-setup` you log out and back in once, because gnome-shell only loads an extension at session start.

## Requirements
- Needs: Python and uv (commands run via uvx)
- Environment variables: `DESKWRIGHT_SESSION`, `DESKWRIGHT_INPUT_BACKEND` — keys and tokens: ask the human for the values, never commit them or put them in shared files.

## Install

### Claude Code (native platform)

**Platform CLI**
```bash
claude mcp add --env DESKWRIGHT_SESSION=<DESKWRIGHT_SESSION> --env DESKWRIGHT_INPUT_BACKEND=<DESKWRIGHT_INPUT_BACKEND> --transport stdio deskwright -- uvx deskwright
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from deskwright <command>

### Codex

**Platform CLI**
```bash
codex mcp add deskwright --env DESKWRIGHT_SESSION=<DESKWRIGHT_SESSION> --env DESKWRIGHT_INPUT_BACKEND=<DESKWRIGHT_INPUT_BACKEND> -- uvx deskwright
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from deskwright <command>

### Cursor

**Config file**
File `.cursor/mcp.json`:
```json
{
    "mcpServers": {
        "deskwright": {
            "command": "uvx",
            "args": [
                "deskwright"
            ],
            "env": {
                "DESKWRIGHT_SESSION": "<DESKWRIGHT_SESSION>",
                "DESKWRIGHT_INPUT_BACKEND": "<DESKWRIGHT_INPUT_BACKEND>"
            }
        }
    }
}
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from deskwright <command>

### GitHub Copilot

**Config file**
File `.vscode/mcp.json`:
```json
{
    "servers": {
        "deskwright": {
            "type": "stdio",
            "command": "uvx",
            "args": [
                "deskwright"
            ],
            "env": {
                "DESKWRIGHT_SESSION": "<DESKWRIGHT_SESSION>",
                "DESKWRIGHT_INPUT_BACKEND": "<DESKWRIGHT_INPUT_BACKEND>"
            }
        }
    }
}
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from deskwright <command>

### Gemini CLI

**Platform CLI**
```bash
gemini mcp add -e DESKWRIGHT_SESSION=<DESKWRIGHT_SESSION> -e DESKWRIGHT_INPUT_BACKEND=<DESKWRIGHT_INPUT_BACKEND> deskwright uvx deskwright
```
Replace <…> with your own values. uvx runs the command named after the package; if the README names another one, use uvx --from deskwright <command>

## When to use it
- open a native app and press a button inside its dialog
- read what another window shows without taking a screenshot
- give the agent its own desktop so it stops stealing the pointer
- record the screen or pull text off it with OCR
- drag an element and rearrange windows

## Example prompts
- Open GIMP and pick New from the File menu
- Start a headless session and open GNOME settings inside it
- Screenshot the window and read the text on it

## Similar
- [REA](https://affpapa.org/en/nevibecode/mcp/rea.md) — MCP server · ★ 386989
- [GitHub](https://affpapa.org/en/nevibecode/mcp/github.md) — MCP server · ★ 33918
- [Netdata](https://affpapa.org/en/nevibecode/mcp/netdata.md) — MCP server · ★ 81115
- [browser-use](https://affpapa.org/en/nevibecode/mcp/browser-use.md) — MCP server · ★ 119160
- [Codebase Memory](https://affpapa.org/en/nevibecode/mcp/codebase-memory.md) — MCP server · ★ 47392
- [mcp-server](https://affpapa.org/en/nevibecode/mcp/mcp-server.md) — MCP server · ★ 43872

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md