\r \r

MCP Scaffolder Skill\r

\r When the user describes an MCP server they want to build, scaffold it completely.\r \r

What to Generate\r

\r

1. Folder structure\r
   • /src/index.ts — server entry point\r
   • /src/tools/ — one file per tool\r
   • /package.json — with @modelcontextprotocol/sdk dependency\r
   • /tsconfig.json — standard TS config\r
   • /.env.example — all required env vars documented\r
   • /README.md — setup instructions\r \r
2. Server entry point (src/index.ts)\r
   • Import and register all tools\r
   • Set transport based on user intent:\r
     • stdio → local tools, file system access, CLI utilities\r
     • HTTP → remote servers, cloud APIs, services\r
   • Include server instructions field — used by Claude Code tool search\r \r
3. Tool definitions\r
   • Each tool gets its own file in /src/tools/\r
   • Include: name, description, inputSchema (zod or JSON Schema)\r
   • Add JSDoc comments explaining what the tool does and when Claude should call it\r \r
4. Claude Code config snippet\r Always output a ready-to-paste config block:\r \r For stdio:\r {\r "mcpServers": {\r "\x3Cserver-name>": {\r "type": "stdio",\r "command": "node",\r "args": ["\x3Cabsolute-path>/build/index.js"],\r "env": { "\x3CENV_VAR>": "your-value-here" }\r }\r }\r }\r \r For HTTP:\r {\r "mcpServers": {\r "\x3Cserver-name>": {\r "type": "http",\r "url": "http://localhost:3000/mcp"\r }\r }\r }\r \r
5. README\r
   • What the server does\r
   • Prerequisites\r
   • All env vars with descriptions\r
   • Build and run instructions\r
   • How to add to Claude Code\r \r

Rules\r

\r

• Never use SSE transport — it is deprecated, use HTTP instead\r
• Always include the server instructions field in the entry point\r
• Keep tool descriptions specific — Claude Code uses these for tool search\r
• Flag any required env vars in .env.example with a comment explaining what they are for\r
• Use TypeScript by default unless user specifies otherwise