learn-claude-code
Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1
learn-claude-code is an educational MCP server and tutorial repository that teaches developers how to build AI agent harnesses from scratch using bash-based tools and progressive lessons. It demonstrates that agency comes from model training, while harnesses provide the operational environment—tools, context management, permissions, and task systems—that enable models like Claude to act effectively. The repository contains 17 progressive chapters, each isolating a single harness mechanism, from basic agent loops to advanced features like multi-agent teams, background task execution, and workflow orchestration.
Key Features
Use Cases
- 01Learning how to build AI agent infrastructure and operational harnesses for LLM-based systems
- 02Understanding the architectural patterns behind coding agents like Claude Code
- 03Implementing tool handlers, permission governance, and context management for production agents
- 04Building multi-agent systems with task coordination and parallel execution capabilities
- 05Studying best practices for agent memory persistence, skill loading, and background task handling
- 06Extending agent capabilities through MCP server integration and custom workflow orchestration
Related MCP Servers
View morepublic-apis
A collective list of free APIs
awesome-python
The definitive list that answers "I want to do X in Python, which tool should I use?"
awesome-selfhosted
A list of Free Software network services and web applications which can be hosted on your own servers
ECC
The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
learn-claude-code — FAQ
What is the learn-claude-code MCP server?+
It's an educational repository and tutorial series that teaches developers how to build agent harnesses—the infrastructure that gives AI models like Claude tools, context, permissions, and task management capabilities. The project emphasizes that intelligence comes from model training, while harnesses provide the operational environment.
How do I install and run learn-claude-code?+
Clone the repository from GitHub, install dependencies with pip install -r requirements.txt, copy .env.example to .env and add your ANTHROPIC_API_KEY, then run individual lesson files like python s01_agent_loop/code.py. Each chapter contains standalone runnable code.
What are the prerequisites for using this MCP server?+
You need Python 3.x installed and an Anthropic API key to run the examples. The lessons are designed for developers who want to understand agent harness engineering, with basic Python and LLM API knowledge recommended.
Which AI clients work with learn-claude-code?+
The tutorial is built around Claude's API and demonstrates patterns applicable to any LLM-based agent system. The s14 MCP plugin chapter shows how to integrate external MCP servers into a unified tool pool compatible with Claude Desktop and similar clients.
Is learn-claude-code free to use?+
Yes, the repository and all tutorial code are open source and free to use. However, running the examples requires an Anthropic API key, which incurs usage costs based on Claude API pricing.
What's the difference between the 17-lesson and legacy 12-lesson tracks?+
The root-level s01-s17 folders are the current canonical version with expanded topics including permissions, hooks, memory, cron scheduling, and goal loops. The legacy docs/ and agents/ folders preserve the older 12-lesson version for existing readers during migration.
How do I install learn-claude-code?+
Open the source repository on GitHub and follow its README. learn-claude-code is a mcp server — MCP Agents Market links you directly to the official repo.
Is learn-claude-code free?+
learn-claude-code is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.