Fifty-Two Ways to Build an Agent Before Lunch

What it does

This repository is a curated cookbook of agent patterns shipped as Jupyter notebooks. It contains 52 tutorials (and growing) organized by skill level and domain—beginner chatbots, LangGraph state machines, MCP tool integrations, academic planners, customer support bots, and more. Each entry is a standalone notebook meant to be opened and executed, not installed as a library.

The interesting bit

Rather than building yet another framework, the author catalogs how to compose existing ones—LangChain, PydanticAI, LangGraph, MCP—across dozens of real-world scenarios. It is unabashedly educational glue code, and that is precisely its value: a pattern library you can break and rebuild.

Key highlights

• 52+ runnable notebooks categorized by difficulty and use case (Beginner, Framework, Educational, Business)
• Covers LangChain, LangGraph, PydanticAI, and the Model Context Protocol (MCP)
• Recently added: LightRAG art tour guide, HR assistant, contextual quoting system, ML/DS assistant
• Accompanied by external blog posts and two companion books on RAG and prompt engineering
• Community-driven with contribution guidelines and an active Discord server

Caveats

• The README is heavily interleaved with book promotions, newsletter signups, and sponsorship banners
• This is a tutorial collection, not a maintained library; expect to copy patterns rather than vendor a dependency

Verdict

Grab this if you are learning agent architectures and want working code to dissect. Skip it if you are hunting for a production-ready framework to drop into your stack—look at the author’s separate Agents Towards Production repo instead.