MCP Agent Loop – v2.0.0

This release marks our first full loop of the Model Context Protocol (MCP) prototype using LLMs.

🔁 Completed Loop: LLM → Tool → LLM

• LLM (planTool.js) chooses which tool to invoke
• Tool executes and logs structured result
• Final LLM response is generated using the tool output (finalReply.js)

🧾 What's Included

✅ v1.0.0 Recap

• Regex-based tool detection with fallback response
• Modular toolRegistry.js with schema-enforced tool cards
• Tools: calculator, time, diceRoll
• CLI test runner and tool introspection (npm run tools)
• Logging via trace.js

🚀 v2.0.0 Additions

• LLM planning via OpenAI GPT-4 (planTool.js)
• Fallback to LLM when no tool matches
• Final response generated by GPT-4 (finalReply.js)
• Full natural-language conversation powered by MCP-style reasoning
• CLI tool planning runner: npm run plan

This version is the first complete simulation of a compliant MCP loop.