Code Execution Mode FAQs

Question 1

How does Code Execution Mode reduce LLM context overhead?

Accepted Answer

Instead of exposing all MCP tools directly to the LLM, which consumes massive context, Code Execution Mode proxies tools into sandboxed code execution. This approach achieves over 95% context overhead reduction, making LLM interactions much more efficient.

Question 2

What are the prerequisites to use Code Execution Mode?

Accepted Answer

You'll need a rootless container runtime (Podman or Docker), `uv` (for project management), and a Python base image (e.g., `python:3.12-slim`). Detailed installation steps are provided in the quick start guide within the README.

Question 3

What is TOON and why is it used?

Accepted Answer

TOON (Token-Oriented Object Notation) is a highly token-efficient format used for bridge responses. It collapses repetitive JSON keys and optimizes arrays, trimming token counts by 30-60% for uniform tables, which helps lower LLM billing and improve processing speed.

Question 4

What security measures are in place for code execution?

Accepted Answer

Code is executed in rootless, network-isolated containers with dropped capabilities, an immutable read-only filesystem, and an unprivileged user. It also enforces resource limits (memory, CPU, PIDs, time) for robust security and auto-cleans temporary IPC directories.

Code Execution Mode

关于

主要功能

使用案例