# anthropic-experimental/sandbox-runtime

A lightweight sandboxing tool for enforcing filesystem and network restrictions on arbitrary processes at the OS level, without requiring a container.

Repository: https://github.com/anthropic-experimental/sandbox-runtime
Canonical: https://ross.abutalabs.com/products/sandbox-runtime
Language: TypeScript
License: Apache-2.0
License Family: permissive
Last push: 2026-08-25T21:52:05+00:00

## Health v2 (maintenance only)
Score: 84/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 22
- inputs: {"age_days": 317, "days_push": 8, "days_rel": 8, "gap_med": 2.0, "n_releases_24m": 37}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5071, forks 415 (observed 2026-08-28T04:09:09.283921+00:00)

## What it is
A lightweight CLI tool and library that enforces filesystem and network restrictions on arbitrary processes at the OS level without containers, using sandbox-exec on macOS and bubblewrap on Linux. It is designed to sandbox AI agents, MCP servers, and bash commands with a secure-by-default allowlist approach.

## Use cases
- sandbox bash commands with filesystem and network restrictions
- run MCP servers with restricted capabilities
- restrict which domains an agent can access over HTTP
- block processes from reading sensitive files like SSH keys
- monitor sandbox violations in real time
- run untrusted code without a container

## When to choose
- you need OS-level sandboxing of arbitrary processes without Docker or containers
- you want to safely run AI agents or MCP servers with least-privilege access
- you need per-command network allowlists and filesystem read/write policies on macOS or Linux

## When to avoid
- you need full container isolation or cross-platform support beyond macOS and Linux
- you require a stable, production-hardened API since this is a beta research preview
- you need sandboxing on Windows or BSD

## Facets
- artifact type: cli-tool
- maturity: experimental
- function: security, networking, file-system, cli, developer-tools
- domain: security, developer-tools
- platform: cli
- tags: sandboxing, os-level-isolation, agent-security, mcp, bubblewrap, sandbox-exec, network-allowlist, filesystem-restrictions, ai-agents, command-line, macos, linux, nodejs

## Member repositories
- anthropic-experimental/sandbox-runtime (main) score 84

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:09.283921+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:02:48.255520+00:00, confidence not recorded.
  - readme: https://github.com/anthropic-experimental/sandbox-runtime (fetched 2026-08-28T04:09:09.283921+00:00, sha d203fdbfc299)
- Data as of 2026-08-30T08:39:29.467469+00:00.