Claude Cookbooks

Anthropic's official collection of runnable Jupyter notebooks, at 51.5k stars and 6.1k forks under an MIT licence, showing how to do specific things with the Claude API rather than explaining them in prose. The repository is organised by capability rather than course order, so it works as a reference you dip into. Classification, retrieval-augmented generation and summarization cover the core capabilities; the tool-use directory holds the most-copied notebooks, including `customer_service_agent.ipynb`, `calculator_tool.ipynb` and `how_to_make_sql_queries.ipynb`; the multimodal directory covers vision end to end, from `getting_started_with_vision.ipynb` through `best_practices_for_vision.ipynb` to `reading_charts_graphs_powerpoints.ipynb` and `how_to_transcribe_text.ipynb`; and third-party examples wire Claude to Pinecone, Voyage AI embeddings and Wikipedia. Directories added as the platform grew, including `claude_agent_sdk`, `patterns/agents`, `skills`, `managed_agents`, `extended_thinking`, `observability`, `tool_evaluation`, `cost_optimization` and `finetuning`, track Anthropic's own agent stack rather than lagging it. Several notebooks are load-bearing for production work rather than demos: `building_evals.ipynb` for automated evaluation, `prompt_caching.ipynb`, `how_to_enable_json_mode.ipynb` for reliable structured output, `building_moderation_filter.ipynb`, and a cost-optimization checklist. Examples are Python-first but the patterns port to any SDK. You need a Claude API key, and Anthropic recommends its Claude API Fundamentals course first if the API is new to you.

Topic

Agentic

Level

Intermediate

Format

Interactive

Cost

Free

Duration

Self-paced; individual notebooks ~20-60 min each

Provider

Anthropic

Hands-on

Yes — code/exercises

Certificate

None

• ✓Call Claude's tool-use API to drive calculators, SQL queries and support agents
• ✓Build a retrieval-augmented pipeline against Pinecone with Voyage AI embeddings
• ✓Handle vision inputs including charts, slide decks and handwritten transcription
• ✓Write automated evaluations for a Claude application with building_evals.ipynb
• ✓Cut token spend using prompt caching and the cost-optimization checklist
• ✓Force reliable structured output with JSON mode and add a moderation filter
• ✓Apply Anthropic's own agent patterns via the claude_agent_sdk and patterns/agents notebooks

• •Maintained by Anthropic itself and updated as the platform ships; the agent SDK, skills and observability directories are recent additions
• •51.5k stars and 6.1k forks, and the cookbook front page drew 343 points and 168 comments on Hacker News
• •Every recipe is an executable notebook you can run and modify, not a snippet in a blog post
• •MIT licensed, so the code can go straight into a commercial project
• •Covers the unglamorous production concerns most tutorials skip: evals, caching, cost, moderation and observability

This page was written from 3 sources, 2 on domains other than github.com.

1. 1.github.com — claude cookbooksvendor
2. 2.news.ycombinator.com — item
3. 3.platform.claude.com — cookbook