MCP Project Initializer

# CLAUDE.md This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository. ## Project Overview This is an MCP (Model Context Protocol) Project Initializer - a command-line MCP server that automates the setup of new AI-powered software projects. The tool acts as a conversational guide through any standard MCP client to set up projects with necessary context, rules, and documentation for AI-assisted development. ## Architecture ### Core Requirements - **Language**: TypeScript - **Runtime**: Node.js (LTS version) - **Protocol**: Model Context Protocol (MCP) communicating over STDIO - **Core Dependency**: `@modelcontextprotocol/sdk` TypeScript SDK - **LLM Interaction**: Uses "Prompt Server / Response" pattern (`mcp_prompt_request` and `mcp_llm_response`) - no direct LLM API keys ### Key Features to Implement 1. **Conversational Project Initialization** - Trigger setup workflow via `/init-mcp` command 2. **Technology-Specific Context Fetching** - Download best-practice rules and SDK docs 3. **Custom Documentation Ingestion** - Handle URLs and raw text input 4. **LLM-Powered PRD Generation** - Generate project documentation using gathered context 5. **Local File System Operations** - Create directories and save files to project structure ## Development Status This is a greenfield project - currently contains only planning documentation: - `docs/prd.md`: Complete Product Requirements Document - `docs/llms-full.txt`: MCP client compatibility matrix and examples ## Key Design Principles - Server must NOT have its own LLM API key - Final script should be executable directly from command line - Focus on TypeScript/Node.js ecosystem initially - Leverage existing MCP protocol patterns for client communication