---
id: agent-sandbox
version: "0.0.30"
license: Apache-2.0
license_treatment: permissive
maintenance: active
---

# agent-sandbox — Python SDK for the All-in-One Sandbox API, >=1.7.0

License: permissive · Maintenance: active · Downloads: 124.7K/mo

## What it is and what it does

Agent Sandbox is a Python SDK that wraps an All-in-One Sandbox API, giving you programmatic access to isolated execution environments for shell commands, file operations, Jupyter notebooks, Node.js scripts, and Model Context Protocol (MCP) interactions. It's designed for agent systems and automation workflows that need to safely execute untrusted or exploratory code in a sandboxed context.

The SDK provides both synchronous and asynchronous interfaces. You connect to a sandbox instance (either locally at http://localhost:8091 or via a cloud provider like Volcengine), then call methods on the client to execute commands, read/write files, run Python in Jupyter kernels, or execute JavaScript. It depends on httpx for HTTP, pydantic for data validation, and volcengine-python-sdk for cloud provider integration.

Use it for:

- Execute shell commands safely in an isolated environment without exposing your host system
- Run untrusted Python code in a Jupyter kernel and capture the output
- Execute JavaScript code in a sandboxed Node.js environment from a Python agent
- Manage files within a sandbox—read, write, and search without direct filesystem access
- Build agent workflows that need to interact with MCP servers in a controlled manner

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

A Python SDK for accessing sandbox, shell, file, Jupyter, Node.js, and MCP services through an All-in-One Sandbox API, with both sync and async support.

Yes, if you need programmatic access to sandboxed execution environments for agent systems or automation. The package is actively maintained, has low install friction, and covers a broad set of execution contexts (shell, Python, Node.js, MCP). Verify whether your use case requires a local sandbox instance or cloud setup, and confirm MCP server availability if you plan to use that feature.

## Install

pip install agent-sandbox
uv add agent-sandbox
poetry add agent-sandbox

## Installing agent-sandbox

Before you install: Low friction install with a pure-Python wheel. Actively maintained as of March 2026. Depends on httpx, pydantic, typing_extensions, and volcengine-python-sdk—all standard, well-maintained packages.

License in practice: Apache-2.0 permissive license allows commercial and private use with minimal restrictions; you must include a copy of the license and state significant changes.

Quickstart:

pip install agent-sandbox

from agent_sandbox import Sandbox

client = Sandbox(base_url="http://localhost:8091")
result = client.shell.exec_command(command="ls -la")
print(result)

Requires a running sandbox server at the specified base_url (e.g., http://localhost:8091 for local development, or Volcengine credentials for cloud deployment).

Verify before relying:

- Whether the Volcengine provider requires cloud credentials to be useful, or if local sandbox mode works standalone
- Performance characteristics and latency for remote sandbox operations
- Whether MCP service integration requires external MCP servers to be running

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 124.7K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

sandbox api client python, shell command execution sdk, jupyter code execution api, nodejs javascript execution, file management sandbox, mcp model context protocol, async sandbox client, sandbox-execution, agent-infrastructure, async-support

[View on SkillFed](https://skillfed.io/packages/agent-sandbox) · [View on PyPI](https://pypi.org/project/agent-sandbox/)