finds.dev

Annyfee/agent-craft

AI Agent 教学仓库 | 系统化 LangChain、RAG、LangGraph、MCP 全栈实战代码 | 万字博客详解 | 开源可运行示例 | 从零构建智能体

View on GitHubHomepage ↗

★ 530PythonMITupdated Oct 2026

// the take

A step-by-step Chinese-language tutorial repo that builds up the modern AI agent stack module by module: raw LLM calls, function calling, LangChain, RAG, LangGraph, MCP servers/clients, and multi-agent patterns, each paired with a CSDN blog post. It's for developers who read Chinese and want a guided path through this stack rather than scattered framework docs.

The module sequence is genuinely well thought out — it goes from bare API calls to Function Calling to LangChain to RAG to LangGraph to MCP to multi-agent orchestration, which mirrors how the ecosystem actually layered itself over the last two years. Each module is self-contained with its own README and runnable scripts, so you can jump into module 8 without having worked through 1-7. It covers MCP server AND client (FastMCP, langchain-mcp-adapters, both stdio and streamable HTTP transports) which most agent tutorials skip entirely. Has actual CI running, which is rare for a teaching repo and suggests the examples are checked, not just written once and left to rot.

All the real explanation lives on external CSDN blog posts, not in the repo — without reading Chinese and visiting another platform, you're left with commented code and no narrative on why decisions were made. Modules 14 (the capstone project combining everything) and 15 (deployment) are explicitly marked as still being written, so the curriculum isn't finished despite being presented as 15 modules. config.py validates that all four API keys (DeepSeek, LangSmith, Amap, ChatGPT) are present at import time regardless of which module you're running, which is annoying friction if you just want to try one example cheaply. This is a learning resource, not something to depend on — no real test suite beyond CI lint/smoke checks, and the code is written for clarity over robustness (no retry logic, minimal error handling), so don't lift any of this directly into production.

We read the README, the directory tree and a key source file in November 2026. Nobody pays to be listed. How we judge →

// at a glance

Stars
530
Forks
75
Language
Python
License
MIT
Last commit
27 Oct 2026

// nearby

If agent-craft isn't quite it.

All repos →