📄 中文摘要
Peter Steinberger 最近在社交媒体上发表的六个字引发了广泛关注，他是 OpenClaw 的创始人，拥有 190,000 个 GitHub 星标，并被 Sam Altman 招募。作者在使用 Ubuntu 多年后，强调了自己日常使用的工具几乎全部为命令行界面（CLI），如 Supabase CLI、Vercel CLI、Docker 和 git 等。尽管去年 MCP 服务器开始流行，作者尝试后发现其不仅占用了 40% 的上下文窗口，还频繁崩溃，并增加了不必要的依赖。因此，CLI 在效率和稳定性方面更具优势，适合构建 AI 代理。文章还提到如何构建自己的 CLI 工具，以提升开发效率。
📄 English Summary
Why CLIs Beat MCP for AI Agents — And How to Build Your Own CLI Army.
Peter Steinberger recently sparked interest with a six-word post on social media, being the creator of OpenClaw with 190,000 GitHub stars and recently recruited by Sam Altman. The author, having built on Ubuntu for years, emphasizes that nearly all the tools used daily are command-line interfaces (CLIs), such as Supabase CLI, Vercel CLI, Docker, and git. Despite the rising trend of MCP servers last year, the author found them to consume 40% of the context window, crash frequently, and add unnecessary dependencies. Therefore, CLIs are more efficient and stable, making them better suited for building AI agents. The article also discusses how to create one’s own CLI tools to enhance development efficiency.