Building an AI Agent That Reads Files and Runs Commands: What the Agent SDK Hands You

When you want an agent that reads and fixes code from one prompt, here's what Anthropic's Agent SDK gives you for free in Python and TypeScript — and when you can skip it.

"I want a program that finds and fixes bugs in our code — do I have to build file reading and command execution myself?" That's where many newcomers stall. Anthropic's Agent SDK is a library that hands you those parts already built.

What it does for you

The docs describe it as using Claude Code as a library: the same tools, agent loop, and context management, programmable in Python and TypeScript. The key point is that built-in tools for reading files, running commands, and editing code mean your agent starts working without you implementing tool execution.

async for message in query(
    prompt="Find and fix the bug in auth.py",
    options=ClaudeAgentOptions(allowed_tools=["Read", "Edit", "Bash"]),
):
    print(message)

Built-in tools include Read, Write, Edit, Bash, Glob, Grep, WebSearch, WebFetch, and AskUserQuestion. A trap for beginners: if you only want to summarize TODOs, allow just ["Read", "Glob", "Grep"] — adding Edit or Bash lets the agent touch files you didn't mean to.

Install snag and when to skip

The Python package needs 3.10+; a No matching distribution found error usually means your interpreter is older. Check with python3 --version. Use the API key method — the docs state claude.ai login isn't allowed for third-party agents without approval.

Skip the SDK when your task is a single text summary or classification with no tools or loops — a plain API call is lighter. Verify the official docs.

Decision to start with: does my task involve reading files or running commands? If yes, allow only the tools you need and run the quickstart.

---

🐦 Faster updates on X: @baegseungh7061
📚 More in this series: Code Practical
💌 Subscribe: Follow on X or grab the RSS