这是一个代理技能，帮助你在任何MCP服务器上添加代码模式——这样LLM就能编写小型处理脚本，在沙箱运行时对应大型API响应，只有脚本的紧凑输出进入上下文窗口。适用于任何支持代理技能的AI代理。

典型结果：65–99%的上下文缩减。

什么是代码模式？

当MCP工具返回大型API响应（例如列出500个Kubernetes播客、200个SCIM用户或数千个GitHub问题）时，整个负载都会进入LLM的上下文窗口——消耗令牌并降低性能。

代码模式则相反：LLM不将原始数据倾倒到上下文中，而是编写一个小型处理脚本。MCP服务器在*沙箱运行时**中对原始数据运行脚本，只有脚本的标准输出进入上下文。

这在知名API（SCIM、Kubernetes、GitHub、Stripe、Slack、AWS等）中尤其有效，因为LLM已经通过训练数据知道响应模式。

灵感来自Cloudflare代码模式和claude上下文模式。

这项技能的作用

这是一项互动式规划技能。它会一步步与你合作，帮助你：

Code Mode Skill

An agent skill that helps you add code mode to any MCP server — so LLMs can write small processing scripts that run against large API responses in a sandboxed runtime, and only the script’s compact output enters the context window. Works with any AI agent that supports agent skills.

Typical result: 65–99% context reduction.

What is Code Mode?

When an MCP tool returns a large API response (e.g. listing 500 Kubernetes pods, 200 SCIM users, or thousands of GitHub issues), that entire payload enters the LLM’s context window — consuming tokens and degrading performance.

Code mode flips the approach: instead of dumping raw data into context, the LLM writes a small processing script. The MCP server runs the script in a sandboxed runtime against the raw data, and only the script’s stdout enters context.

This works especially well with well-known APIs (SCIM, Kubernetes, GitHub, Stripe, Slack, AWS, etc.) because LLMs already know the response schemas from training data.

Inspired by Cloudflare Code Mode and claude-context-mode.

What This Skill Does

This is an interactive planning skill. It works with you step-by-step to:

1. Understand your MCP server (language, framework, which tools return large data)
2. Select a sandbox that fits your server language and security requirements
3. Plan the implementation together
4. Implement the code mode tool, sandbox executor, and wiring
5. Benchmark before/after context sizes to verify the reduction

Supported Server Languages

All sandbox options are isolated from the host filesystem and network by default.

Install

npx skills add chenhunghan/code-mode-skill

Or manually copy the SKILL.md and references/ directory into your agent’s skills directory.

Usage

Just ask your agent to add code mode to your MCP server:

• “I want to add code mode to my MCP server to reduce token usage”
• “Add a sandboxed code execution tool to my TypeScript MCP server”
• “My get-users tool returns 50KB+ responses, help me reduce context size”

The skill will guide you through the process interactively.

License

MIT