Скачать или смотреть Claude Code Tool Use Testing

Cloud Code lets you use tool calling, also known as tool use or function calling, which is super handy for making AI agents more reliable and useful. Basically, instead of the AI just giving you random text, you can force it to answer in a specific format like JSON, making it way easier to automate tasks or extract exact info from documents. Tool use has actually been around for years, but not a lot of people use it, which is a shame because it is really powerful for pulling out structured data, like checking if a contract has certain details or getting exact values out of prompts.

In Cloud Code, you pick from supported models like Sonnet 4, and then for each tool you set up, you define what parameters it takes and what the required output should look like, like grabbing the weather for a city in a certain unit. You can even force the model to use the tool, and if there's missing info, it'll either ask for it or fail, depending on your setup. Setting it up with the Cloud Code Python SDK is pretty direct, and you can have the model help you write most of the boilerplate if you want.

This setup is great for tasks like fetching info from databases, running checklists on long documents, or pulling out specific fields users mention in their prompts, and you can do it today with free open source models like GPT-OSS, not just on pricey APIs. If you want to automate business tasks, cut down on manual data entry, or make sure your AI always gives you structured output, tool use is the way to go. There are some limits, like context window size, but you can work around that by processing big documents in chunks, and always be careful about permissions if you let your AI change data.