Week 14: Tool Use — Teaching Claude to Do Real Things

Last week your agent had mocked tools. This week you give it real ones.

The key idea: Claude doesn't run tools. Claude asks you to run them. Your code runs the tool and sends the result back. Claude just picks which tool to call and what arguments to pass.

How Tool Schemas Work

When you call Claude with tools, you send a JSON schema for each tool. The schema tells Claude:

name: What to call it
description: When to use it (Claude reads this to decide)
input_schema: What arguments it needs (JSON Schema format)

Claude reads these descriptions and decides which tool fits the situation. Better descriptions = better tool selection.

How the Flow Works

You: "What's on the front page of Hacker News?"
 |
 v
Claude: "I should use web_search to check Hacker News"
        → tool_use: web_search(url="https://news.ycombinator.com")
 |
 v
Your code: fetches the URL, returns the HTML text
 |
 v
Claude: "Here are the top stories: ..."

The stop_reason in Claude's response tells you what happened:

"end_turn" = Claude is done talking, here's the answer
"tool_use" = Claude wants to use a tool, run it and send the result back

This Week's Code

Open tools.py — this has three real tool functions:

web_search: Fetches a URL and returns the text content
read_file: Reads a local file
calculate: Evaluates math expressions

Open tool_agent.py — this is the agent that uses those tools in a loop.

Setup

Same as last week. Make sure you have your .env file:

cp .env.example .env

Install dependencies:

pip install anthropic python-dotenv

Run It

python tool_agent.py

Try these:

"Read the file tools.py and tell me what functions are in it"
"What is 210 + 310?"
"Fetch https://httpbin.org/json and summarize what you get"

Tasks

Run the agent with the examples above. Watch how Claude picks tools.
Read tools.py. Understand each function. What could go wrong with each one?
Add a new tool. Ideas:
- list_directory(path) — lists files in a folder using os.listdir()
- write_file(path, content) — writes content to a file
- get_weather(city) — fetch weather from a free API like wttr.in: https://wttr.in/{city}?format=3
Improve the descriptions. Change a tool's description and see if Claude uses it differently.
Fill in reflection.md.

Deliverables

tool_agent.py and tools.py with at least one new tool
reflection.md filled in
Copy-paste of one run showing the agent using multiple tools

Stuck?

Same pattern as always — paste your error into Claude or ChatGPT:

"I'm building a tool-use agent with the Anthropic SDK. My tool returns [X] but Claude doesn't seem to understand it. Here's my tool definition: [paste]. What's wrong?"

README.md