LLM SandboxBLOCK
Lightweight and portable LLM sandbox runtime (code interpreter) Python library.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
LLM Sandbox
Securely Execute LLM-Generated Code with Ease
[](https://sonarcloud.io/summary/newcode?id=vndeellm-sandbox)
[](https://sonarcloud.io/summary/newcode?id=vndeellm-sandbox) [](https://pypi.org/project/llm-sandbox/) [](https://img.shields.io/github/v/release/vndee/llm-sandbox) [](https://github.com/vndee/llm-sandbox/actions/workflows/main.yml?query=branch%3Amain) [](https://www.codefactor.io/repository/github/vndee/llm-sandbox) [](https://codecov.io/gh/vndee/llm-sandbox) [](https://doi.org/10.5281/zenodo.21760525) [](https://deepwiki.com/vndee/llm-sandbox)
LLM Sandbox is a lightweight and portable sandbox environment designed to run Large Language Model (LLM) generated code in a safe and isolated mode. It provides a secure execution environment for AI-generated code while offering flexibility in container backends and comprehensive language support, simplifying the process of running code generated by LLMs.
Documentation: https://vndee.github.io/llm-sandbox/
✨ New: This project now supports the [Model Context Protocol (MCP)](https://vndee.gi
5bda1b473e27OBSERVED · 2026-09-25Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control. Replace the environment placeholders with a token scoped to the least it needs.
claude mcp add llm-sandbox -- uvx llm-sandbox==0.3.43
Exposed tools (3)
2 read · 1 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
execute_code | write | Execute code in a secure sandbox environment and automatic visualization capture. |
get_language_details | read | Get the details of a language. |
get_supported_languages | read | Get the list of supported languages. |
Trust audit
BLOCKgrade D · trust 67/100 Do not install this without reading the findings. The audit found something that could harm you or your machine.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | WARN |
| L2 | Instruction surface (what it tells the agent) | FAIL |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (2 observation(s))
- Network
- declared (1 observation(s))
- Shell
- declared (1 observation(s))
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (25)
# IMPORTANT: When using readOnlyRootFilesystem, you MUST mount
# IMPORTANT: When using readOnlyRootFilesystem, you MUST also provide
module = importlib.import_module(name)
.bandit
.pre-commit-config.yaml
.sonarcloud.properties
.zenodo.json
"import pickle\ndata = pickle.loads(b'malicious_data')",
description="Inject malicious code through eval() function",
description="Execute system commands through exec() function",
weak_hash = hashlib.md5(password.encode() + salt).hexdigest()
weak_hash = hashlib.md5(password.encode() + salt).hexdigest()
weak_hash = hashlib.md5(password.encode() + salt).hexdigest()
print(f'Secret: {secret}')"import os\nfilename = '../../../etc/passwd'\nwith open(filename, 'r') as f:\n print(f.read())",
session.copy_to_runtime(temp_file.name, "../../etc/shadow")
self.mixin._validate_container_path("../../etc/shadow")"network_mode": "none", # no egress: injected code cannot exfiltrate or fetch a second stage
"network_mode": "none", # no egress: injected code cannot exfiltrate or fetch a second stage
"network_mode": "none", # no egress: injected code cannot exfiltrate or fetch a second stage
"network_mode": "none", # no egress: injected code cannot exfiltrate or fetch a second stage
"network_mode": "none", # no egress: injected code cannot exfiltrate or fetch a second stage
plot_path.write_bytes(base64.b64decode(plot.content_base64))
f.write(base64.b64decode(plot.content_base64))
f.write(base64.b64decode(plot.content_base64))
Gates applied: no_behavioural_pass.
5bda1b473e27full audit observations/trust-audit/mcp-server/vndee__llm-sandbox-1.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-09-25 | 5bda1b473e27 | BLOCK | D | 67 | first audit |
Questions
What is the LLM Sandbox MCP server?
Lightweight and portable LLM sandbox runtime (code interpreter) Python library.
What tools does LLM Sandbox expose?
3 in total: 2 read-only, 1 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is LLM Sandbox safe to connect to an agent?
No — not without reading the findings first. The audit graded it D (67/100) and found 2 critical or high issues in the source. Each one is listed on this page with the file and line it is on.
What credentials does LLM Sandbox need?
It reads SECRET_KEY from the environment. Give it a token scoped to the least it needs — an agent that can be talked into calling a tool can be talked into calling it with your credentials.
How does LLM Sandbox run?
It speaks stdio, so it runs as a local process your client starts. It is published on PyPI as llm-sandbox.
How current is this page?
The grade is for one exact copy of the source (5bda1b473e27), read on 2026-09-25. The repository is watched and re-audited when it changes.