LLM Sandbox runs on your own machine — the client starts it, so there is no endpoint to ping. 93 684 installs a week from pypi. Last commit 14 Sep 2026.
Securely run LLM-generated code in isolated containers across 7 languages and 3 container backends.
We read the source, 18 h ago · rules 3dff92dd89df
What this server is able to do. For an MCP server this is often the job itself — a terminal server runs commands because that is what it is for. Listed so you know what you are plugging in, not as an accusation.
этот файл ставится пользователю, но в репозитории его нет
- name: Update CITATION.cff to the released version
env:
RELEASE_VERSION: ${{ github.event.release.tag_name }}
Is this your server and something here is wrong? Tell us — corrections are free and do not require a plan.
We found places where it runs commands, builds paths or queries from values it is given. None of that is a flaw by itself — it becomes one when the code changes, and code changes quietly between releases. We re-read it on every one.
This server runs on your own machine — install it with the package manager and the client starts it for you. Package name taken from the official registry entry.
claude mcp add llm-sandbox -- uvx llm-sandbox
{
"mcpServers": {
"llm-sandbox": {
"args": [
"llm-sandbox"
],
"command": "uvx"
}
}
}
[mcp_servers.llm-sandbox]
command = "uvx"
args = ["llm-sandbox"]
{
"mcpServers": {
"llm-sandbox": {
"args": [
"llm-sandbox"
],
"command": "uvx"
}
}
}
{
"mcpServers": {
"llm-sandbox": {
"args": [
"llm-sandbox"
],
"command": "uvx"
}
}
}
This one needs environment variables set before it will start:
BACKEND (Container backend to use. Must match the installed extra: mcp-docker, mcp-podman, or mcp-k8s.), DOCKER_HOST (Docker or Podman socket URL, e.g. unix:///var/run/docker.sock), KUBECONFIG (Path to kubeconfig file when BACKEND=kubernetes.), NAMESPACE (Kubernetes namespace used for sandbox pods when BACKEND=kubernetes.), COMMIT_CONTAINER (Commit the container after a run so installed libraries persist between sessions.), KEEP_TEMPLATE (Keep the base image after the session ends to avoid re-pulling it on the next run.), SANDBOX_NETWORK_MODE (Network mode for the sandbox container. Set to 'none' for hardened isolation. Docker and Podman backends only.), SANDBOX_READ_ONLY (Mount the sandbox root filesystem read-only. Recommended: true. Docker and Podman backends only.), SANDBOX_CAP_DROP (Comma-separated Linux capabilities to drop. Recommended: ALL. Docker and Podman backends only.), SANDBOX_SECURITY_OPT (Comma-separated container security options, e.g. no-new-privileges. Docker and Podman backends only.), SANDBOX_MEMORY (Memory limit for the sandbox container, e.g. 4g. Docker and Podman backends only.), SANDBOX_CPUS (Fractional CPU allocation for the sandbox container, e.g. 1.5. Docker and Podman backends only.).
The author declared them in the registry entry; get the values from the project itself.
Run AI-generated code safely. gVisor-isolated Python/JS sandboxes for any MCP client. Pay per run.
Build Apps and run code in 30 languages — sandboxed, with persistent sessions for agent loops.
Execute code in isolated cloud containers. Supports Python, JavaScript, Bash, and Go.
Orchestrates language servers into 65 code-intelligence tools across 31 languages, token-optimized.
Code generation from JSON recipes, linked type graphs, and API schemas across 11 languages.
Create and manage secure sandboxes for running AI agent code
Execute code in 8 languages (Python, JS, TS, Go, Java, C++, C, Bash) in gVisor sandboxes.
Run Python code in a secure sandbox without local setup. Declare inline dependencies and execute s…
Answers built from our own checks of this server.