Claude Cookbooks (recettes officielles Anthropic)
Anthropic's official repo (41,000+ stars) of executable notebooks and recipes: RAG, tool use, sub-agents, multi-model orchestration, full agents, with the reference implementations of the 'Building Effective Agents' essay. Not a course, a toolbox: when your agent hits a wall, you come here for the pattern that works, written and maintained by the people building the model. It is also where the advisor and orchestrator notebooks live, the ones that turn an expensive model into a manager of cheaper ones.
Strengths
- Reference implementations maintained by Anthropic, aligned with the actual state of the API
- Covers the agent patterns that matter: RAG, tool use, sub-agents, orchestration, evals
- Executable notebooks: you run, you see the behavior, you adapt
Limitations
- Python + notebooks: the ideas are universal, the raw code does not fit every stack
- A toolbox, not a path: without a precise goal, you get lost in it
Best for
- A dev or technical PM building their first real agent who wants to start from a proven pattern
- A team wanting to settle an agent architecture debate with the official reference