LLM Automation Cookbook — copy-paste Claude (Anthropic) recipes in Python

Short, self-contained, copy-paste recipes for automating real work with the Claude (Anthropic) API in Python. Every recipe is one file you can drop into your project — structured output, tool use, streaming, prompt caching, batching, retries, vision, and more. All recipes are linted in CI to use current models and avoid deprecated parameters.

Each file is standalone. Copy the one you need, set ANTHROPIC_API_KEY, run it.

Recipes

#	Recipe	What it shows
01	Call the Claude API in Python	The minimal request/response
02	Use a system prompt	Steer tone and format
03	Extract structured JSON with Pydantic	messages.parse → typed object
04	Tool use / function calling loop	The agentic tool-use loop
05	Stream a response	Token-by-token output
06	Adaptive thinking + effort	Opus 4.8 reasoning controls
07	Prompt caching	Cut cost on repeated context
08	Count tokens accurately	count_tokens, not tiktoken
09	Retry on rate limits (429)	Exponential backoff
10	Batch many prompts (50% off)	Message Batches API
11	Analyze an image (vision)	Image input
12	Fast text classification with Haiku	Cheap, single-word labels

Run a recipe

git clone https://github.com/skycandykey1/llm-automation-cookbook
cd llm-automation-cookbook
pip install -r requirements.txt
export ANTHROPIC_API_KEY=sk-ant-...        # Windows: setx ANTHROPIC_API_KEY ...
python recipes/01_basic_message.py

Bonus: tested helpers

cookbook/utils.py ships two small, unit-tested helpers you can reuse:

• estimate_cost(input_tokens, output_tokens, model) — USD cost from token counts
• backoff_delay(attempt) — exponential backoff schedule for retries

Tests

Everything is verified offline — no API key, no network:

pip install -r requirements-dev.txt
python -m pytest -q

• test_utils.py — unit tests for the helpers
• test_recipes_lint.py — lints every recipe: approved model id, imports the SDK, has a docstring, and uses no deprecated params (temperature, budget_tokens, ...) or non-Anthropic libraries

License

MIT — see LICENSE.

---

💼 Built by skycandykey1 — available for AI agent & automation contract work. I build agents, MCP servers, and LLM automation. → skycandykey456@gmail.com · https://github.com/skycandykey1