Manufact：在 Codex/Claude Code 里测试你的 MCP 服务器
用 Manufact 仪表盘给 MCP 服务器加测试套件，让 Codex、Claude Code、Pi 等真实 CLI 跑任务并由 LLM 裁判打分，附修复建议。
how to test your MCP server in codex, claude code, pi, etc: 1. open your server in the manufact dashboard → Testing → Add test suite. write the task the way a user would ask for it, ideally one that needs several tools in order 2. set the bar and pick clients. an LLM judge scores each run 0-100 against your rubric (70 passes by default) add Claude Code, Codex CLI and Pi. if you have an MCP App, add ChatGPT or Claude too, since the CLIs don't render UI 3. run it. each agent gets a fresh sandbox with the real CLI pointed at your server. the judge scores from a separate sandbox and never calls your server 4. 4. read the report and fix. each server bug comes with evidence and a Fix button that writes the prompt for your coding agent run more than one agent. in our example codex retried past two bugs and passed, claude failed try it 👇