π
mcp
MCP server for AI agent sandboxes β exposes sandbox tools over Model Context Protocol
0 installs
Trust: 39 β Low
Data
Installation
npx @agent-sandbox/mcpAsk AI about mcp
Powered by Claude Β· Grounded in docs
I know everything about mcp. Ask me about installation, configuration, usage, or troubleshooting.
0/500
Loading tools...
Reviews
Documentation
Agent Sandbox
Agent Sandbox gives AI agents an isolated sandbox backed by Docker β somewhere to run code, install packages, execute commands, and manage files without touching your machine.
Features
- Isolated by default β each sandbox is a Docker container, fully separated from the host
- Persistent storage β every sandbox gets a dedicated
/datavolume that survives across commands - Self-hosted β runs on your own machine via Docker, no cloud or API keys required
Primitives
create delete list read write upload download
Interfaces
| Package | Role |
|---|---|
@agent-sandbox/cli | Human-facing command line wrapper over the core primitives |
@agent-sandbox/mcp | MCP server that exposes the primitives as tools |
@agent-sandbox/api | Small programmatic wrapper for app/server integration |
Quick Start
CLI
# install or upgrade
npm install -g @agent-sandbox/cli
agent-sandbox create
agent-sandbox list
MCP
{
"mcpServers": {
"agent-sandbox": {
"command": "npx",
"args": ["-y", "@agent-sandbox/mcp"]
}
}
}
API
npm install @agent-sandbox/api
import { AgentSandbox } from "@agent-sandbox/api";
const sandbox = new AgentSandbox();
const created = await sandbox.create();
await sandbox.write({
sandboxId: created.sandboxId,
input: "printf 'hello\\n' > /proc/1/fd/1",
});
console.log(
(
await sandbox.read({
sandboxId: created.sandboxId,
})
).output,
);
await sandbox.delete({
sandboxId: created.sandboxId,
});
Documentation
Use the package READMEs for package-specific usage and reference:
Skill
npx skills add https://github.com/usamaasfar/agent-sandbox/tree/main/skills/agent-sandbox-cli
