langchain-sandrpod

SandrPod sandbox integration for Deep Agents.

Installation

pip install langchain-sandrpod

Quick Start

from langchain_sandrpod import SandrPodClient
from deepagents import create_deep_agent
from deepagents.middleware import FilesystemMiddleware

client = SandrPodClient(api_url="http://localhost:8080")

with client.sandbox("my-sandbox") as sb:
    # Run a command directly
    result = sb.execute("python3 -c 'print(42)'")
    print(result.output)  # 42

    # Or plug the sandbox into a deepagents agent
    agent = create_deep_agent(
        middleware=[FilesystemMiddleware(backend=sb)]
    )
    result = agent.invoke({"messages": [{"role": "user", "content": "List the files under /workspace"}]})

Sandbox surface

See examples/ — including 04_personal_mcp.py, which exposes the personal MCP tools on an employee machine to a remote agent via the sandbox's /mcp endpoint.

Environment Variables