Analysis updated 2026-06-24
Build a Python script that uses Claude to automatically review or refactor code files in a repository
Define a custom tool so Claude can query your own database or call your API during a session
Add a hook that blocks Claude from running certain shell commands in automated pipelines
Run multi-turn conversations with Claude from a script without opening a browser
| anthropics/claude-agent-sdk-python | autoscrape-labs/pydoll | google-research/arxiv-latex-cleaner | |
|---|---|---|---|
| Stars | 6,844 | 6,841 | 6,838 |
| Language | Python | Python | Python |
| Setup difficulty | easy | easy | easy |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | developer | developer | researcher |
Figures from each repo's GitHub metadata at analysis time.
Requires an active Claude Code installation, bundled in the package so no separate install needed.
The Claude Agent SDK for Python is a library that lets developers write Python programs that talk to Claude Code, the AI coding tool from Anthropic. Instead of interacting with Claude through a web interface, this SDK allows a Python script to send prompts and receive responses automatically, making it possible to build tools that use Claude as a component. The simplest use is the query function, which sends a prompt and returns an asynchronous stream of response messages. You can set a system prompt, limit how many turns the conversation runs, and specify which of Claude's built-in tools (such as reading files, writing files, or running shell commands) are automatically approved versus which require confirmation. For longer interactive sessions, there is a client class called ClaudeSDKClient. The client supports custom tools, which are Python functions you define and offer to Claude. When Claude decides it needs to use one, it calls your function directly within the same running program rather than starting a separate process. This avoids the overhead and complexity of managing external tool servers. Hooks are another feature of the client. A hook is a Python function that runs at specific points in the agent loop, such as just before Claude uses a tool. Hooks can inspect what Claude is about to do, block certain actions, or inject feedback. The example in the README shows a hook that blocks any shell command containing a particular pattern. The SDK bundles the Claude Code command-line tool, so no separate installation is needed. The package requires Python 3.10 or newer and is installed through pip, the standard Python package manager. The full README is longer than what was shown.
A Python library that lets your code talk to Claude Code programmatically, send prompts, stream responses, define custom tools, and add hooks that control what the AI is allowed to do.
Mainly Python. The stack also includes Python, asyncio.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.