BentoRun
Enables AI agents to execute Python code securely within isolated environments.
BentoRun is a Model Context Protocol (MCP) server designed to empower AI agents with secure and isolated Python code execution capabilities. It leverages NsJail, a lightweight process isolation tool by Google, to create robust sandboxes for each execution session, whether in Docker or Google Cloud Run. This ensures that AI agents can run Python code safely without risk of interfering with the host system, while still providing controlled access to external resources like the internet and necessary libraries.
Key Features
01Controlled network access for external API calls and package installations.
02Utilizes Linux namespaces, cgroups, and seccomp-bpf for robust isolation.
03Secure Python code execution within isolated sandboxes using NsJail.
04Configurable sandbox environment with custom dependencies and system mounts.
050 GitHub stars
06Strict resource limits (memory, CPU, output size) and filesystem isolation.
Use Cases
01Provide sandboxed environments for untrusted code execution in cloud or containerized setups.
02Enable AI agents to safely execute Python code.
03Integrate as an MCP server with tools like Gemini CLI for agent tool use.