yucai0302

Legacy2mcp — AI skill for Claude Code

AI community

Turn legacy business systems into MCP / Tools that AI can call — an Agent Skill for Claude Code / Codex / OpenCode / CodeBuddy / Cursor / Trae.

How to install Legacy2mcp

This entry records only its repository, not the path inside it, so there is no exact command to give. Open yucai0302/legacy2mcp and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Legacy2mcp does

Turn legacy business systems into MCP / Tools that AI can call — an Agent Skill for Claude Code / Codex / OpenCode / CodeBuddy / Cursor / Trae.

Alternatives in AI

Graphify — AI coding assistant skill (Claude Code, Codex, OpenCode, OpenClaw, Factory Droid, Trae) 20.9k ★
AI Guide — 程序员鱼皮的 AI 资源大全 + Vibe Coding 零基础教程，分享 OpenClaw 保姆级教程、大模型玩法（DeepSeek / GPT / Gemini / Claude）、最新 AI 资讯、Prompt 提 11.8k ★
Rb Setup — First-time setup 1.3k ★

README

legacy2mcp

**[English](README.md)** | **[简体中文](README.zh-CN.md)**

**Turn hand-crafted legacy business systems into MCP / Tools that AI can call fluently.**

An Agent Skill that plugs into mainstream coding agents (Claude Code / Codex / OpenCode / CodeBuddy / Cursor / Trae / Qoder …). It reads a legacy repository, understands business functions **module by module**, produces a visual review page where every finding carries its source-code provenance, and — after human approval — generates a standalone MCP adapter project.

Survey → Module Map (human picks scope) → Capability Extraction (APIs / IO / logic / flows)
       → Visual Review (human approves) → Generate MCP Server → Verify & Deliver

Why

The common bottleneck in enterprise AI adoption: AI is smart, but it can't reach those systems that have been running for years. Every system needs a "retrofit" — wrapping business capabilities into MCP/Tools. This skill turns the retrofit into a **verifiable, traceable, human-in-the-loop** pipeline:

Never swallows the whole repo: legacy systems run hundreds of thousands of lines — dumping them all into an LLM is unrealistic. Build a module map first, let a human pick the scope, then read module by module. Optionally plug in CodeGraph / Serena for further token savings (fully local, optional).
Every conclusion has a source: each capability carries file:lines + symbol name. Fabricated APIs and logic are forbidden.
Two human gates: module scope, and capability list + exposure method — both confirmed by a human before the pipeline moves on.
Visual review: a self-contained single-file HTML (interaction inspired by Archify) — module tree, capability cards, input/output tables, logic checklists, business flow diagrams, exposure selection. Click "Export" to get the retrofit plan JSON.
AI proposes, human decides: MCP stdio / MCP HTTP / REST+OpenAPI / in-process tools — the AI recommends per a deci