CrateBay

Open-source local AI sandbox. Run code safely on your machine — no cloud, no cost.

CrateBay gives any AI agent (Claude, Cursor, Windsurf, your own) a secure sandbox to execute code, install packages, and manage files — all running locally inside a lightweight VM. No Docker installation required.

Why CrateBay?

AI agents need a safe place to run code. Cloud sandboxes (E2B, Modal) charge per minute and send your code off-machine. CrateBay runs everything locally:

• Zero cost — no cloud bills, no usage limits
• Private — code never leaves your machine
• Fast — local VM, no network round-trip
• Works with any AI — MCP protocol, works with Claude Desktop, Cursor, Windsurf, and any MCP-compatible client
• No Docker required — built-in VM runtime (macOS: Virtualization.framework, Linux: KVM, Windows: WSL2)

How It Works

Your AI Agent                    CrateBay                         Local VM
(Claude, Cursor, etc.)           (MCP Server)                     (Docker inside VM)
       │                              │                                │
       │  "run this Python script"    │                                │
       ├─────────────────────────────►│                                │
       │                              │  create sandbox + exec code    │
       │                              ├───────────────────────────────►│
       │                              │                                │
       │                              │  stdout/stderr + exit code     │
       │       result                 │◄───────────────────────────────┤
       │◄─────────────────────────────┤                                │

Quick Start

1. Install

# macOS (Apple Silicon & Intel)
brew install --cask cratebay

# Or download from Releases

2. Connect to Your AI

Add to your MCP client config (e.g. Claude Desktop claude_desktop_config.json):

{
  "mcpServers": {
    "cratebay": {
      "command": "cratebay-mcp"
    }
  }
}

3. Use It

Tell your AI:

"Create a Python sandbox and run: print('Hello from CrateBay')"

CrateBay handles the rest — VM startup, container creation, code execution, result delivery.

Features

MCP Server — Let Any AI Run Code

The cratebay-mcp binary exposes sandbox tools via the Model Context Protocol:

Tool	What It Does
sandbox_run_code	Create sandbox + execute code + return result (one-shot)
sandbox_create	Create a persistent sandbox from template
sandbox_exec	Run a command in an existing sandbox
sandbox_install	Install packages (pip, npm, apt)
sandbox_upload / sandbox_download	Transfer files in/out of sandbox
sandbox_list	List running sandboxes
sandbox_stop / sandbox_delete	Lifecycle management

Desktop App — Visual Sandbox Management

The CrateBay desktop app provides:

• Chat interface — talk to an AI assistant that manages sandboxes through natural language
• Sandbox dashboard — see running sandboxes, resource usage, logs
• Image management — search, pull, and manage container images
• MCP server management — connect external MCP tool servers
• Settings — LLM provider config, runtime settings, registry mirrors

CLI — Headless Operations

cratebay sandbox create --template python-dev
cratebay sandbox exec <id> -- python -c "print('hello')"
cratebay sandbox list
cratebay sandbox stop <id>

Pre-built Sandbox Templates

Template	Image	Use Case
python-dev	Python 3.12 + pip	Data analysis, scripting, ML
node-dev	Node.js 20 + npm	Web development, scripting
rust-dev	Rust stable + cargo	Systems programming
ubuntu-base	Ubuntu 24.04	General purpose

Architecture

┌─────────────────────────────────────────────────────┐
│  CrateBay                                            │
│                                                      │
│  ┌──────────────┐  ┌──────────┐  ┌───────────────┐  │
│  │ cratebay-mcp │  │ GUI App  │  │ cratebay-cli  │  │
│  │ (MCP Server) │  │ (Tauri)  │  │ (CLI)         │  │
│  └──────┬───────┘  └────┬─────┘  └──────┬────────┘  │
│         └───────────────┼───────────────┘            │
│                         │                            │
│              ┌──────────▼──────────┐                 │
│              │   cratebay-core     │                 │
│              │   (Rust library)    │                 │
│              └──────────┬──────────┘                 │
│                         │                            │
│              ┌──────────▼──────────┐                 │
│              │  Built-in Runtime   │                 │
│              │  macOS: VZ.framework│                 │
│              │  Linux: KVM/QEMU   │                 │
│              │  Windows: WSL2     │                 │
│              │       ↓            │                 │
│              │  Docker in VM      │                 │
│              └─────────────────────┘                 │
└─────────────────────────────────────────────────────┘

Tech stack: Tauri v2 | React 19 | Rust | bollard | SQLite | pi-agent-core

Compared To

	CrateBay	E2B	Docker Desktop
Runs locally	Yes	No (cloud)	Yes
AI-native (MCP)	Yes	API only	No
Cost	Free	$0.01/min	Free / $5+/mo
Privacy	Code stays local	Code on cloud	Code stays local
No Docker required	Yes (built-in VM)	N/A	Requires Docker
Open source	MIT	Partial	No

Status

v0.9.0 — Core sandbox infrastructure complete, working toward v1.0 release.

See docs/progress.md for detailed development status and docs/ROADMAP.md for the release plan.

Contributing

This project uses AGENTS.md for AI-assisted development. See docs/ for technical specs and workflow guides.

License

MIT