The idea
Once you have read Build an MCP Server, the steps are mechanical — and mechanical work is exactly what an AI is good at. The Scaffolder is itself an MCP server. You add it to your AI client once; from then on you can ask your client to generate a brand-new MCP server for any task, and the Scaffolder hands back complete, runnable code.
What it exposes
scaffold_mcp_server— give it a task description (and optionally a language and transport); it returns a full project:package.json, source,README, and a list of any TODOs to finish.explain_mcp_concept— ask how any part of MCP works and get a developer-focused explanation.
@modelcontextprotocol/server).Install it
You need Node.js 20+ and an Anthropic API key.
cd mcp-server
npm install
npm run buildRegister it with Claude Desktop
Add it to claude_desktop_config.json — the client launches it as a subprocess over stdio:
{
"mcpServers": {
"scaffolder": {
"command": "node",
"args": ["/absolute/path/to/mcp-server/dist/index.js"],
"env": { "ANTHROPIC_API_KEY": "sk-ant-..." }
}
}
}Or with Claude Code
claude mcp add scaffolder --env ANTHROPIC_API_KEY=sk-ant-... \
-- node /absolute/path/to/mcp-server/dist/index.jsUse it
Restart your client and just ask. For example:
"Scaffold an MCP server that searches a Postgres
database of products and returns matches as JSON."The client calls scaffold_mcp_server, and you get a complete project back in the conversation — review the TODOs, drop in your credentials, build, and connect it like any other MCP server.