Nano Agent

概要

Nano Agent serves as a Master Control Program (MCP) server designed to manage and orchestrate experimental, small-scale engineering agents. Its primary purpose is to facilitate the testing and comparison of agentic capabilities across various cloud-based and local Large Language Models (LLMs), evaluating them based on performance, speed, and cost. It moves beyond single prompt calls, focusing on how well agents can chain multiple tools to achieve real engineering results, providing crucial insights into the practical application and efficiency of different LLM providers and models.

主な機能

• Designed for experimentation with robust testing, error handling, and token usage tracking
• Utilizes a nested agent architecture where the MCP server spawns internal agents for task execution
• Provides a unified interface, allowing all LLM providers to be accessed through the same OpenAI SDK
• Enables agents to perform file system operations like reading, writing, editing, and analyzing files
• 55 GitHub stars
• Supports multiple LLM providers including OpenAI (GPT-5), Anthropic (Claude), and Ollama (local models)

ユースケース

• Delegating complex engineering tasks to agents and scaling up compute through integration with MCP clients like Claude Code.
• Testing and comparing the agentic capabilities of various cloud and local LLMs (e.g., GPT-5 vs. Claude vs. Ollama) across performance, speed, and cost metrics.
• Systematically evaluating multiple LLM models in parallel using Higher Order Prompt (HOP) and Lower Order Prompt (LOP) patterns to benchmark results.