Exploring Early Indicators of AGI in Coding Agents

Exploring Early Indicators of AGI in Coding Agents: A Case Study on MCP-Powered Systems

This study, conducted under the COTRUGLI Business School Co-lab initiative in Q1 and Q2 2025, investigates whether Model Context Protocols (MCPs) can enhance large language models (LLMs) to exhibit early indicators of Artificial General Intelligence (AGI) in coding agents. We developed a Retrieval-Augmented Generation (RAG) Software-as-a-Service (SaaS) application using Streamlit, integrating n8n workflows and Supabase for authentication and coupon management. The Cline Coding Agent, powered by Grok 3 Mini and augmented by five MCPs (Context7, Sequential Thinking, Knowledge Graph Memory, GitHub, and Supabase), completed 90% of the application in nine days for approximately $30 in API costs, compared to a non-MCP baseline that failed within 48 hours. These findings demonstrate that MCPs significantly enhance LLMs’ planning, reasoning, and contextual awareness, approximating early AGI-like behaviours. However, challenges in complex debugging highlight the gap to true AGI. This paper, the first in a series on Context Engineering, underscores the transformative potential of MCPs in AI-driven software development and sets the stage for future experiments with flagship models and local LLMs.

Keywords: AGI, AI, MCP, LLM, coding agent, artificial intelligence, intelligence, reasoning, open-source, technology, Cline, Grok, Claude

AGI AI Coding Agent