taw-computer

<p align="center"> <img src="https://img.shields.io/badge/🖥️_taw--computer-Give_any_AI_a_real_computer-blue?style=for-the-badge&labelColor=000" alt="taw-computer" height="40"> </p>

<p align="center"> <strong>Your AI writes code. This lets it <em>use</em> a computer.</strong> </p>

<p align="center"> <a href="https://github.com/the-agents-work/taw-computer/stargazers"><img src="https://img.shields.io/github/stars/the-agents-work/taw-computer?style=flat-square&logo=github&color=yellow" alt="Stars"></a>  <a href="https://github.com/the-agents-work/taw-computer/blob/main/LICENSE"><img src="https://img.shields.io/github/license/the-agents-work/taw-computer?style=flat-square&color=green" alt="MIT License"></a>  <a href="https://github.com/the-agents-work/taw-computer/issues"><img src="https://img.shields.io/github/issues/the-agents-work/taw-computer?style=flat-square" alt="Issues"></a>  <a href="#quick-start"><img src="https://img.shields.io/badge/get_started-5_minutes-brightgreen?style=flat-square" alt="Get Started"></a> </p>

<p align="center"> <a href="#quick-start">Quick Start</a> · <a href="#demo">Demo</a> · <a href="https://shipkit.cc">Hosted Version</a> · <a href="CONTRIBUTING.md">Contributing</a> · <a href="https://github.com/the-agents-work/taw-computer/issues">Report Bug</a> </p>

---

What if your AI could do everything you do on a computer?

Not just write code — but open a browser, click buttons, fill forms, run servers, test in real browsers, install anything, and see the screen?

taw-computer is an open-source MCP server that gives AI agents a full Ubuntu desktop inside Docker. Your AI connects, gets a real computer, and works like a human would.

No internal LLM. No chat UI. Your AI is the brain. This is the body.

<br>

Demo

<!-- 🎬 Replace with your actual demo GIF/video --> <!-- Record: start Claude Code → "build me a landing page" → AI creates VM, codes, opens browser, shows result --> <!-- Recommended: use asciinema for terminal, or screen record VNC + terminal side by side -->

<p align="center"> <em>📹 Demo coming soon — <a href="https://github.com/the-agents-work/taw-computer/stargazers">star this repo</a> to get notified!</em> </p>

<!-- When ready, uncomment: <p align="center"> <img src="docs/demo.gif" alt="taw-computer demo — AI builds a website from scratch" width="800"> <br> _{AI builds a full website from a single prompt — writing code, installing packages, and testing in a real browser.} </p> -->

<br>

Why taw-computer?

Other tools let AI write code. taw-computer lets AI use a computer.

	ChatGPT / Claude	Cursor / Copilot	Lovable / Bolt	taw-computer
Write code	✅	✅	✅	✅
Run shell commands	❌	Limited	Sandboxed	Full Ubuntu
Browse the web	❌	❌	❌	Real Chromium
See & click the screen	❌	❌	❌	Desktop + VNC
Install any software	❌	❌	❌	apt/npm/pip
Test in real browser	❌	❌	Preview only	Playwright + CDP
Persist across sessions	❌	❌	✅	Snapshots
Self-hostable	❌	❌	❌	100% yours

<br>

Quick start

Get running in under 5 minutes:

# 1. Clone & build
git clone https://github.com/the-agents-work/taw-computer.git
cd taw-computer
docker build -f images/Dockerfile.taw -t taw-computer-base .

# 2. Install & start
npm install && npm start

Then add to your AI client:

<details open> <summary><strong>Claude Code</strong> (~/.claude/mcp.json)</summary>

{
  "mcpServers": {
    "taw-computer": {
      "command": "npx",
      "args": ["tsx", "/path/to/taw-computer/mcp/index.ts"]
    }
  }
}

</details>

<details> <summary><strong>Cursor</strong></summary>

Add to Cursor MCP settings (Settings → MCP Servers) — same JSON format as above.

</details>

<details> <summary><strong>Claude Desktop</strong></summary>

Add to claude_desktop_config.json — same JSON format as above.

</details>

<details> <summary><strong>Any MCP client</strong></summary>

taw-computer speaks standard MCP over stdio. Any client that supports MCP can connect.

</details>

<details> <summary><strong>Remote server (SSH)</strong> — run on a beefy machine, use from your laptop</summary>

Got a powerful server / Mac Mini / VPS? Run taw-computer there and connect from anywhere:

{
  "mcpServers": {
    "taw-computer": {
      "command": "ssh",
      "args": ["user@your-server", "cd /path/to/taw-computer && npx tsx mcp/index.ts"]
    }
  }
}

Your laptop (Claude Code)
    ↕ SSH (stdin/stdout piped over network)
Remote server (taw-computer + Docker)
    ↕ Docker
Ubuntu sandbox

Setup:

1. On the server: install Docker, clone repo, build image, npm install
2. On the server: enable SSH (sudo systemctl enable ssh)
3. On your laptop: ssh-copy-id user@your-server (passwordless login)
4. Add the MCP config above — done!

Watch via VNC: open http://your-server:6080 in your browser.

</details>

That's it. Now tell your AI: "Create a VM and build me a website" — and watch it work.

<br>

What can it do?

🖥️ "Build me a landing page"

AI creates a VM → scaffolds Next.js → writes components → starts dev server → opens browser to check → iterates until it looks right

🌐 "Go to Amazon and find the best laptop under $1000"

AI opens Chromium → navigates to Amazon → searches → scrolls → extracts prices → compares → reports back

🧪 "Run E2E tests on my deployed app"

AI launches Playwright → navigates to your URL → fills forms → clicks buttons → asserts results → reports failures

🔧 "Set up a PostgreSQL database with sample data"

AI runs apt install postgresql → creates database → writes seed script → runs it → verifies with queries

📸 "What does my app look like on mobile?"

AI takes desktop screenshot → resizes viewport → screenshots again → compares → suggests CSS fixes

<br>

How it works

┌─────────────────────────────────────────────────────┐
│  Your AI Client                                     │
│  Claude Code · Cursor · Claude Desktop · any MCP    │
└───────────────────────┬─────────────────────────────┘
                        │ MCP protocol (stdio)
┌───────────────────────▼─────────────────────────────┐
│  taw-computer MCP server          30+ tools         │
│  vm · shell · files · browser · desktop · search    │
└───────────────────────┬─────────────────────────────┘
                        │ Docker API
┌───────────────────────▼─────────────────────────────┐
│  Ubuntu 22.04 Sandbox              isolated container│
│                                                      │
│   bash    Chromium + CDP    xfce4 Desktop + VNC     │
│     git npm pip curl          Playwright              │
│       python node              xdotool scrot          │
│                                                      │
│   /workspace ← your project files live here          │
└──────────────────────────────────────────────────────┘

<br>

30+ tools

<details open> <summary><strong>VM Management</strong> — create, destroy, snapshot, resume</summary>

Tool	What it does
vm_create	Spin up a new sandbox. Returns VNC URL to watch live
vm_destroy	Destroy (auto-saves snapshot for later)
vm_reset	Destroy + delete snapshot (fresh start)
vm_restart	Restart container, keep all files
vm_status	CPU, RAM, disk, uptime, top processes
vm_list	List running sandboxes
vm_rename	Rename a VM
snapshot_list	List saved snapshots
snapshot_delete	Delete a snapshot

</details>

<details> <summary><strong>Shell & Files</strong> — full Ubuntu command line + file ops</summary>

Tool	What it does
exec	Run any command: git, npm, pip, curl, docker, anything
fs_read	Read a file
fs_write	Write a file (creates parent dirs)
fs_edit	Find-and-replace in a file
fs_list	ls / recursive find
fs_search	grep for patterns
code_search	ripgrep with regex, file types, context
file_upload	Upload file into VM (base64, max 50MB)

</details>

<details> <summary><strong>Browser (CDP/Playwright)</strong> — real browser, not a simulator</summary>

Tool	What it does
browser_navigate	Go to URL, wait for load
browser_snapshot	Screenshot + numbered overlays on every clickable element
browser_click_ref	Click element #N from snapshot
browser_type_ref	Type into element #N
browser_extract	Read page text (CSS selector or full page)
browser_eval	Run JavaScript in page
browser_wait_for	Wait for selector / text / network idle
browser_console_logs	Read console.log, console.error, etc.
browser_network_errors	Catch 404s, CORS errors, failed requests
browser_run_test	Run a Playwright test script
browser_open	Open Chrome via desktop (fallback)
browser_close	Kill Chrome
web_search	Google search → top 8 results

</details>

<details> <summary><strong>Desktop</strong> — see and control the GUI</summary>

Tool	What it does
desktop_screenshot	JPEG screenshot of the whole desktop
desktop_click	Click at (x, y)
desktop_type	Type text into focused window
desktop_key	Key combos: ctrl+c, alt+tab, Return, etc.
desktop_scroll	Scroll up/down
desktop_drag	Drag from A to B

</details>

<br>

Set-of-Mark: how browser automation actually works

Most "computer use" tools guess pixel coordinates. We use Set-of-Mark prompting — the AI sees numbered badges on every interactive element:

Step 1: browser_snapshot
        → AI sees screenshot with [1] Login  [2] Search  [3] Cart  ...

Step 2: browser_click_ref(ref=2)
        → clicks the Search box precisely

Step 3: browser_type_ref(ref=2, text="laptop", submit=true)  
        → types and presses Enter

Step 4: browser_snapshot
        → sees new page with results [4] [5] [6] ...

No coordinate guessing. No CSS selector fragility. The AI sees what it's clicking.

<br>

VNC — watch your AI work in real time

Every sandbox comes with a noVNC web viewer. Open the URL in your browser and watch:

• 🖱️ AI navigating websites and clicking buttons
• ⌨️ AI writing code in the terminal
• 🏗️ AI building and testing applications
• 🐛 AI debugging by inspecting the screen

Perfect for demos, debugging, and building trust in AI agents.

<br>

What's inside each sandbox

	Included
OS	Ubuntu 22.04
Desktop	xfce4 + Xvfb + x11vnc + noVNC
Browser	Playwright Chromium (native arm64 + amd64)
Languages	Node.js 20, Python 3, build-essential
CLI	git, curl, wget, jq, ripgrep, tree, nano, vim
DB clients	PostgreSQL, MariaDB, Redis
Dev tools	GitHub CLI, yq, httpie
Automation	xdotool, scrot, imagemagick, xclip

<br>

Configuration

Variable	Default	Description
MAX_SANDBOXES	3	Max concurrent VMs
SANDBOX_TYPE	auto	auto / docker / firecracker
DOCKER_IMAGE	taw-computer-base	Base image
DOCKER_MEMORY_MB	4096	RAM per container
DOCKER_CPUS	2	CPUs per container
DESKTOP_RESOLUTION	1280x720	Screen resolution

<br>

Requirements

	Minimum
Docker	Docker Desktop or Docker Engine
Node.js	20+
RAM	~4GB per sandbox
Disk	~5GB for base image

<br>

Project structure

taw-computer/
├── mcp/
│   ├── index.ts            # MCP server — stdio, 30+ tool handlers
│   └── browser.ts          # Playwright CDP + Set-of-Mark engine
├── sandbox/
│   ├── SandboxManager.ts   # Abstract interface
│   ├── DockerSandbox.ts    # Docker implementation
│   ├── FirecrackerSandbox.ts # Firecracker microVM (optional)
│   ├── NetworkManager.ts   # Network isolation
│   ├── config.ts           # Env-based config
│   └── index.ts            # Auto-detect backend
├── images/
│   └── Dockerfile.taw      # Ubuntu sandbox image
├── .github/
│   ├── workflows/ci.yml    # CI: typecheck + Docker build
│   └── ISSUE_TEMPLATE/     # Bug report + feature request
├── package.json
├── CONTRIBUTING.md
└── LICENSE (MIT)

<br>

Contributing

We'd love your help! See CONTRIBUTING.md.

Ideas for first contributions:

• 🎨 Record a demo GIF for this README
• 📝 Write a tutorial ("Build X with taw-computer")
• 🔧 Add a new MCP tool (audio? clipboard? multi-tab?)
• 🐳 Build a slimmer Docker image
• 🧪 Add automated tests
• 📦 Support Podman / containerd

<br>

Hosted version

Don't want to self-host? shipkit.cc — managed taw-computer with:

• Chat UI (just type what you want)
• Auth & team collaboration
• One-click app sharing
• No Docker setup needed

<br>

FAQ

<details> <summary><strong>How is this different from Lovable / Bolt / v0?</strong></summary>

Those are closed-source, hosted-only products that generate code. taw-computer gives AI a real computer — it can run servers, browse the web, install anything, and interact with any desktop app. It's also open source and self-hostable.

</details>

<details> <summary><strong>How is this different from OpenInterpreter / Open Hands?</strong></summary>

OpenInterpreter runs code on your local machine (risky). Open Hands uses its own LLM orchestration. taw-computer is just the computer — no built-in LLM, no opinions about orchestration. Your existing AI client (Claude Code, Cursor, etc.) is the brain. taw-computer is a pure MCP server.

</details>

<details> <summary><strong>Is it safe? Can the AI break my system?</strong></summary>

Each sandbox is an isolated Docker container with its own filesystem, network, and process space. Nothing inside can touch your host system. Containers have memory/CPU/PID limits. When you're done, destroy the VM.

</details>

<details> <summary><strong>Can I use it with GPT-4 / Gemini / local models?</strong></summary>

Yes — any AI client that supports MCP can connect. The server doesn't care which LLM is behind the client.

</details>

<details> <summary><strong>Does it work on Mac / Windows / Linux?</strong></summary>

Yes. Anywhere Docker runs, taw-computer runs. The sandbox image supports both arm64 (Apple Silicon) and amd64 (Intel/AMD).

</details>

<br>

Star History

<p align="center"> <a href="https://star-history.com/#the-agents-work/taw-computer&Date"> <img src="https://api.star-history.com/svg?repos=the-agents-work/taw-computer&type=Date" alt="Star History" width="600"> </a> </p>

<p align="center"> If taw-computer is useful to you, <a href="https://github.com/the-agents-work/taw-computer/stargazers"><strong>give it a ⭐</strong></a> — it helps others find it. </p>

<br>

Related

• Model Context Protocol — the protocol taw-computer speaks
• Claude Code — AI coding CLI that supports MCP
• Playwright — browser automation powering CDP tools

License

MIT — do whatever you want with it.