Claude Cookbooks
Anthropic's collection of 150+ working recipes — agent patterns, tool use, RAG, evals, caching and cost optimisation — as runnable notebooks rather than prose.
The gap between “I read the docs” and “I have this working” is usually a runnable example. The Cookbook is 150+ of them, maintained by Anthropic and open to contributions.
What is in it
| Category | Covers |
|---|---|
| Agent patterns | Multi-agent systems, orchestration, workflows |
| Tools | Tool use, programmatic tool calling, tool search with embeddings |
| Managed Agents | Production deployment patterns for Managed Agents |
| Agent SDK | Building agentic applications on the SDK |
| RAG and retrieval | Knowledge graphs, text-to-SQL, retrieval systems |
| Multimodal | Vision, image analysis, document processing |
| Skills | Excel, PowerPoint and PDF automation |
| Evals | Evaluation and testing patterns |
| Integrations | Third-party services |
Recent additions cover cost optimisation, extended thinking, prompt caching and observability.
Why it is often the better first stop
Two categories in particular answer questions this reference gets asked and cannot fully answer:
- Evals. “Should I use Sonnet 5 or Opus 5?” has no general answer — it has your answer, which comes from running an evaluation on your own tasks. The eval recipes are how you build one.
- Cost optimisation. Caching and batching arithmetic is easy to describe and easy to get wrong. A working notebook settles it.
Cookbook versus course
Building with the Claude API is nine structured hours that teach the platform. The Cookbook is a reference you dip into when you already know what you are trying to build. Different tools; most people end up using both.
Related concepts
Messages API
GAThe core Claude API endpoint — you send a list of messages and a model ID, you get a response, and every other platform feature is a parameter on top of it.
Platform
Tool use
GAThe mechanism by which Claude calls functions you define, or server-side tools Anthropic hosts, and uses the results to continue its work.
Agent Stack
Agent SDK
GAThe SDK for building custom agents on Claude Code's engine — the same tools, permissions, skills and hooks, with your own orchestration around them.
Agent Stack
Prompt caching
GAReusing already-processed parts of a prompt across requests, at a tenth of the input price on a cache hit.
Platform
Building with the Claude API
GAAnthropic Academy's flagship developer course — 67 lessons, 8 quizzes and 9 hours covering prompting, tool use, RAG, agents, MCP and production patterns.
Academy
Sources
- 01Claude CookbookOfficialplatform.claude.com
- 02anthropics/claude-cookbooks on GitHubOfficialgithub.com