Claude API Cookbook

A comprehensive collection of Python examples, patterns, and implementations for building applications with the Anthropic Claude API. This repository serves as both a learning resource and a reference guide for developers working with Claude.

Overview

This cookbook covers the full spectrum of Claude API capabilities, from basic prompting to advanced agent architectures. Each file is a standalone example with detailed documentation, best practices, and production-ready patterns.

Contents

Core Capabilities

• Prompting.py - Prompt evaluation framework with automated dataset generation, testing, and grading
• Tool_use.py - Tool integration patterns including datetime tools, reminders, and batch operations
• Tool_streaming.py - Real-time streaming with tool use
• Tools_multi-turn.py - Multi-turn conversations with tool execution
• Structured_data.py - Extracting structured data from Claude responses
• StructuredData_Exercise.py - Practical exercises for structured data extraction
• Thinking.py - Extended thinking capabilities for deeper reasoning tasks
• Images.py - Image analysis, OCR, multi-image comparison, and vision capabilities
• PDF.py - PDF processing and analysis
• Citations.py - Grounding responses with citations and references

Managed Agents

A new execution model: persistent, session-based agents running in sandboxed cloud environments. Each file is self-contained and demonstrates one capability of the Managed Agents API (research preview).

• managed_agents/README.md - Concept map, decision guide, and client.beta.* namespace reference
• managed_agents/01_multi_agent_coordinator.py - Coordinator + sub-agent roster, parallelization, specialization, and escalation patterns
• managed_agents/02_outcomes_with_rubric.py - Outcome-driven sessions: define a rubric, let the agent iterate until it satisfies it
• managed_agents/03_dreams_memory_consolidation.py - Async memory consolidation: deduplicate and resolve contradictions across sessions
• managed_agents/04_webhooks.py - Flask webhook handler with signature verification and idempotency
• managed_agents/05_advisor_strategy.py - Tiered Opus / Sonnet / Haiku model routing for cost-efficient multi-agent systems

Advanced Features

• RAG_system.py - Complete RAG (Retrieval-Augmented Generation) implementation with:

  • Vector and BM25 hybrid search
  • Contextual enrichment
  • Reranking strategies
  • Multiple chunking approaches

• Web_search.py - Web search integration and information retrieval

• Caching.py - Prompt caching for cost optimization and performance

• Code_execution.py - Safe code execution capabilities

• MCP.py - Model Context Protocol implementation guide for building servers and clients

• Agents_and_Workflows.py - Architectural patterns:

  • Workflows vs Agents decision framework
  • Parallelization patterns
  • Chaining strategies
  • Routing workflows

Evaluation & Testing

• Running_Eval.py - Evaluation framework for testing prompts
• Test_Datasets.py - Dataset generation and management
• Prompt_Dataset.py - Prompt testing datasets
• Code_Grading.py - Automated code evaluation
• Grader_Exercise.py - Grading system exercises
• Model_Grading.py - Model output evaluation

Utilities & Demos

• streaming_demo.py - Streaming response demonstrations
• prefill_and_stop_demo.py - Prefill and stop sequence examples
• stop_sequences_template.py - Stop sequence patterns
• structured_data_template.py - Templates for structured output
• test_client.py - Client testing utilities
• Text_editor_tool.py - Text editing tool implementation
• Claude_Code.py - Claude Code integration examples

Prerequisites

pip install anthropic python-dotenv voyageai

For Managed Agents webhook examples:

pip install flask

For MCP examples:

pip install "mcp[cli]>=1.8.0"

Setup

1. Create a .env file in the project root:

ANTHROPIC_API_KEY=your_api_key_here
VOYAGE_API_KEY=your_voyage_key_here  # For RAG examples

2. Each Python file is self-contained and can be run independently:

python Prompting.py
python Tool_use.py
python RAG_system.py

Key Features

🎯 Production-Ready Patterns

Each example includes error handling, best practices, and considerations for production deployment.

📚 Comprehensive Documentation

Inline documentation explains not just how to use features, but when and why to use them.

🔧 Modular Design

Helper functions and reusable components make it easy to integrate patterns into your own projects.

🚀 Progressive Learning

Examples progress from basic concepts to advanced architectures, building on earlier patterns.

Architecture Patterns

Workflows

• Parallelization - Split complex evaluations into focused, parallel tasks
• Chaining - Sequential processing for multi-step refinement
• Routing - Direct requests to specialized pipelines

Agents

• Tool Design - Abstract, composable tools for maximum flexibility
• Environment Inspection - Verify actions and adapt behavior
• Multi-Turn Planning - Complex task decomposition

Use Cases

This cookbook demonstrates implementations for:

• 💬 Conversational AI applications
• 📊 Data extraction and analysis
• 🔍 Semantic search and RAG systems
• 🤖 Autonomous agents and workflows
• 📝 Document processing and OCR
• 🧮 Complex reasoning and problem-solving
• 🔧 Tool integration and API orchestration

Claude's Contribution

This repository was developed with significant contributions from Claude Code (Anthropic) in collaboration with ArchieCur.

Original cookbook (2025): powered by Claude Sonnet 4.5 (claude-sonnet-4-5-20250929). Claude Code assisted in:

• Architecture Design - Designing modular, extensible patterns for each capability
• Code Implementation - Writing production-ready Python code with proper error handling and type hints
• Documentation - Creating comprehensive inline documentation and pattern explanations
• Best Practices - Incorporating lessons learned from real-world Claude API usage
• Educational Structure - Organizing examples progressively from basic to advanced concepts
• Testing & Validation - Ensuring examples are practical and production-ready

Managed Agents update (May 2026): powered by Claude Sonnet 4.6 (claude-sonnet-4-6). Added the managed_agents/ folder covering the Managed Agents API research preview — six self-contained files spanning multi-agent coordination, outcome-driven sessions, Dreams memory consolidation, webhooks, and the advisor strategy pattern.

The collaboration between human guidance and Claude's technical capabilities resulted in a resource that bridges theoretical understanding with practical implementation.

Best Practices

🔒 Security

• Never commit API keys (use .env files)
• Validate and sanitize user inputs
• Implement rate limiting for production use

💰 Cost Optimization

• Use prompt caching for repeated patterns
• Choose appropriate models (Haiku for simple tasks, Sonnet for balance, Opus for quality)
• Batch similar requests when possible
• Monitor token usage

⚡ Performance

• Implement streaming for better UX
• Use async/await for concurrent operations
• Cache embeddings and frequent computations
• Optimize image sizes before sending

🎯 Reliability

• Start with workflows before resorting to agents
• Implement proper error handling and retries
• Test prompts systematically with evaluation frameworks
• Monitor and log for continuous improvement

Examples in Action

Simple Tool Use

from anthropic import Anthropic

client = Anthropic()
response = client.messages.create(
    model="claude-sonnet-4-5",
    max_tokens=1024,
    tools=[get_current_datetime_schema],
    messages=[{"role": "user", "content": "What time is it?"}]
)

RAG System

from RAG_system import RAGSystem

rag = RAGSystem(
    chunking_strategy="sentences",
    use_contextual_enrichment=True,
    use_reranking=True
)

rag.add_documents(documents)
answer = rag.query("Your question here", k=3)

Extended Thinking

response = chat(
    messages,
    thinking=True,
    thinking_budget=2048
)

Contributing

This repository is designed as a learning resource. Feel free to:

• Use these patterns in your own projects
• Adapt examples to your specific needs
• Share feedback or suggestions
• Report issues or inconsistencies

Resources

• Anthropic Documentation
• Claude API Reference
• Model Context Protocol

License

This is an educational resource. Use these examples as reference for your own projects.

---

Built with Claude - Demonstrating the power of human-AI collaboration in creating comprehensive technical resources.