Ghost OS

v1.0.0Browser Automationstable

Full computer-use for AI agents. Self-learning workflows. Native macOS. No screenshots required.

accessibilityai-agentsautomationclaude-codecomputer-use
Share:
1,462
Stars
0
Downloads
0
Weekly
0/5

What is Ghost OS?

Ghost OS is a Model Context Protocol (MCP) server that allows AI assistants like Claude, Cursor, and VS Code to full computer-use for ai agents. self-learning workflows. native macos. no screenshots required.

Full computer-use for AI agents. Self-learning workflows. Native macOS. No screenshots required.

This server falls under the Browser Automation category on MCPgee, the world's largest MCP server directory with 33,000+ servers.

Features

  • Full computer-use for AI agents. Self-learning workflows. Na

Use Cases

Computer control for AI agents
Self-learning workflows on native macOS
ghostwright

Maintainer

LicenseMIT
Languageswift
Versionv1.0.0
UpdatedMay 21, 2026
Statushealthy
Maintenanceactive

Works with

ClaudeOpenAIwindowsmacoslinux

Installation

Manual Installation

npx ghost-os

Configuration

Configuration Details

Config File

claude_desktop_config.json

Performance

Response Metrics

Response Time< 200ms
ThroughputMedium

Resource Usage

Memory UsageLow
CPU UsageLow

How to Set Up and Use Ghost OS

Ghost OS is a native macOS computer-use framework for AI agents that lets any MCP-compatible AI see and control every application on your Mac without relying on screenshots. It reads the macOS Accessibility tree for structured, labeled data about every UI element in every app, falls back to a local ShowUI-2B vision model only when needed, and exposes 29 tools to an AI agent via the MCP protocol. Its self-learning recipe system lets a frontier model watch you perform a workflow once, synthesize it into a parameterized JSON recipe, and then replay that workflow reliably using a small local model — eliminating repeated expensive reasoning for recurring tasks.

Prerequisites

  • macOS 14 or later (Apple Silicon recommended for local vision model performance)
  • Homebrew package manager installed
  • An MCP-compatible AI client such as Claude Code, Cursor, or VS Code with an MCP extension
  • Input Monitoring permission granted in System Settings > Privacy & Security (required for self-learning recipes)
  • Accessibility permission granted to the MCP client application
1

Install Ghost OS via Homebrew

Install Ghost OS using the official Homebrew tap. This installs the ghost binary, the ghost-vision vision sidecar, pre-built recipes, and the MCP configuration file.

brew install ghostwright/ghost-os/ghost-os
2

Run the setup wizard

Run 'ghost setup' to configure macOS permissions (Accessibility and Input Monitoring), register Ghost OS as an MCP server in your AI client, install built-in recipes, and download the ShowUI-2B local vision model.

ghost setup
3

Grant required macOS permissions

When prompted by the setup wizard, grant Accessibility permission (System Settings > Privacy & Security > Accessibility) to your AI client, and Input Monitoring permission (same location) to enable the self-learning recipe system.

4

Verify MCP connection in your AI client

Open Claude Code or your chosen MCP client and confirm Ghost OS appears as a connected server. You should see the 29 Ghost OS tools available, including ghost_run, ghost_click, ghost_type, and ghost_learn_start.

5

Run your first recipe

Ask your AI agent to execute a built-in recipe such as downloading a paper from arXiv. The agent calls ghost_run with the recipe name and parameters — Ghost OS handles all clicking, typing, and navigation.

ghost_run recipe:"arxiv-download" params:{"query":"chain of thought prompting"}
6

Teach Ghost OS a new workflow

Use ghost_learn_start to begin recording a workflow, perform the task manually, then call ghost_learn_stop. Ghost OS synthesizes your actions into a reusable, parameterized recipe.

# In your AI client:
# 1. Agent calls: ghost_learn_start task_description:"send email in Gmail"
# 2. You perform the task manually
# 3. Agent calls: ghost_learn_stop
# 4. Agent calls: ghost_recipe_save

Ghost OS Examples

Client configuration

After running 'ghost setup', Ghost OS automatically registers itself. The resulting MCP config entry looks like this:

{
  "mcpServers": {
    "ghost-os": {
      "command": "ghost",
      "args": ["mcp"],
      "env": {}
    }
  }
}

Prompts to try

These prompts demonstrate what Ghost OS enables in any MCP-connected AI client.

- "Send an email to [email protected] with subject 'Q4 Report' and a brief summary of today's metrics."
- "Download the latest arxiv paper on chain-of-thought prompting to my Desktop."
- "Open Slack and send 'standup done' to the #team channel."
- "Watch me fill out this expense form, then create a recipe so you can do it for me next time."
- "Take a screenshot of the current app and describe what you see on screen."

Troubleshooting Ghost OS

Homebrew install fails with an Xcode version error on macOS beta

Use the manual install instead: download the latest release tarball from github.com/ghostwright/ghost-os/releases, extract the binaries, copy them to /opt/homebrew/bin/, copy the recipe and GHOST-MCP.md files to /opt/homebrew/share/ghost-os/, then run 'ghost setup'.

Ghost OS cannot click elements in a web app or dynamic content area

Ghost OS falls back to the local ShowUI-2B vision model when the macOS Accessibility tree lacks detail. Ensure the vision sidecar downloaded successfully during 'ghost setup'. If not, re-run 'ghost setup' with a stable internet connection.

Self-learning (ghost_learn_start) does not capture any actions

Input Monitoring permission must be granted to the ghost binary. Go to System Settings > Privacy & Security > Input Monitoring and add or re-enable the ghost-os entry, then restart your AI client.

Frequently Asked Questions about Ghost OS

What is Ghost OS?

Ghost OS is a Model Context Protocol (MCP) server that full computer-use for ai agents. self-learning workflows. native macos. no screenshots required. It connects AI assistants to external tools and data sources through a standardized interface.

How do I install Ghost OS?

Follow the installation instructions on the Ghost OS GitHub repository. Clone the repo, install dependencies, and add the server config to your AI client.

Which AI clients work with Ghost OS?

Ghost OS works with all major MCP-compatible AI clients including Claude Desktop, Claude Code, Cursor, VS Code (GitHub Copilot), Windsurf, and Cline.

Is Ghost OS free to use?

Yes, Ghost OS is open source and available under the MIT license. You can use it freely in both personal and commercial projects.

Browse More Browser Automation MCP Servers

Explore all browser automation servers available in the MCPgee directory. Each server includes setup guides for Claude, Cursor, and VS Code.

Quick Config Preview

{ "mcpServers": { "ghost-os": { "command": "npx", "args": ["-y", "ghost-os"] } } }

Add this to your claude_desktop_config.json or .cursor/mcp.json

Read the full setup guide →

Ready to use Ghost OS?

Browse our complete directory of 33,000+ MCP servers, read setup guides for your editor, and start building with the Model Context Protocol.

33,000+ ServersFree & Open SourceStep-by-Step Guides