GhostDesk MCP server

Give any AI agent a full desktop — it sees the screen, clicks, types, and runs apps like a human. Automate anything with a UI: browsers, legacy software, internal tools. No API needed. One Docker command.

What GhostDesk does

Summarised from the project README; check the repository for the latest details.

Tools exposed (17)

• Screen
• screen_shot – `region?` — `{x, y, width, height}`, cropped at native resolution · `format` `"webp"` \
• mouse_move – `x` int · `y` int · `region?` — the `screenshot` region these were read off
• mouse_click – `x` · `y` · `button` `"left"` \
• mouse_double_click – `x` · `y` · `button` (`left`) · `region?`
• mouse_drag – `fromx` · `fromy` · `tox` · `toy` · `button` (`left`) · `region?`
• mouse_scroll – `x` · `y` · `direction` `"up"` \
• key_type – `text` string — Unicode, newlines and tabs, layout-independent
• key_press – `keys` string — one chord, `+` between tokens. Modifiers: `ctrl`/`control`, `alt`/`option`, `shift`, `super`/`meta`/`win`/`cmd`/`command`. N
• Clipboard
• clipboard_get – —
• clipboard_set – `text` string
• Apps
• app_list – —
• app_running – —
• app_launch – `command` string — an `exec` from `applist`, arguments not accepted · `waitforwindow` bool (`true`)
• app_status – `pid` int — one returned by `applaunch` · `lines` int (`50`)

Requirements

• Text + vision — the agent perceives the desktop through screenshots and
• Tool use — GhostDesk exposes its tools as function calls; the model must
• MCP client — the host needs to speak Streamable HTTP MCP to reach the
• WebP image support — GhostDesk returns screenshots as WebP by default to
• YV17labs/llama-cpp-webp —

How to install GhostDesk

Listed on Socketlist – add this badge to your README:

Badge markdown[](https://socketlist.pages.dev/s/yv17labs-ghostdesk/)

Maintain this project? Get it featured for free during launch →

Compare with chrome-devtools-mcp →

Related Browser Automation MCP servers