luochang212/dive-into-langgraph

What it solves

LangGraph 1.0 is a powerful agent framework, but learning it from scattered official docs is slow and overwhelming. This project provides a structured, hands-on tutorial that walks developers from the first ReAct agent to advanced topics like state graphs, memory, human-in-the-loop, MCP, parallelization, and RAG. It also removes the need to memorize the framework: the tutorial is packaged as an Agent Skill for Claude Code, so an AI assistant can write high-quality LangChain and LangGraph code directly.

How it works

The project is an open-source eBook built as a series of 14 Jupyter notebooks, each focusing on one major capability. The chapters progress from quickstart to production-style patterns, with examples for building agents, adding middleware, managing memory, integrating MCP servers, implementing supervisor patterns, and more. It also includes a runnable Gradio app (Chapter 13) and a debugging page via langgraph dev. As of March 2026, the tutorial has been converted into an Agent Skill: you can install it into Claude Code with npx skills add luochang212/dive-into-langgraph and then have Claude Code use it to generate accurate LangGraph/LangChain code.

Who it's for

This is for developers who want to learn LangGraph 1.0 quickly and practically—especially those building AI agents with LangChain and LangGraph. It's also useful for anyone who wants to use Claude Code as a coding assistant for LangChain/LangGraph projects, since the Skill gives the AI direct access to the tutorial's knowledge.

Highlights

• Entirely based on LangGraph 1.0, with no legacy v0.6 remnants.
• 14 chapters covering state graphs, memory, human-in-the-loop, MCP, supervisor patterns, parallelization, RAG, web search, and more.
• Includes a working Gradio streaming agent app and a debugging page built on langgraph-cli.
• Now available as an Agent Skill for Claude Code—install via npx skills add luochang212/dive-into-langgraph.
• Chinese-first with an English translation available.