Basic agent with the Claude Agent SDK

This example runs the agent loop behind Claude Code as a library, using the Claude Agent SDK. The options set the harness defaults before the loop starts: the system prompt, the tools the agent may use, the permission mode that governs approvals, and the maximum number of turns.

Requirements

• Python 3.10+
• An Anthropic API key set as an environment variable (ANTHROPIC_API_KEY)
• The Claude Code CLI available on the PATH, since the SDK runs it as a subprocess (npm install -g @anthropic-ai/claude-code)

Steps for running this example in the shell

1. Install dependencies:

python -m venv .venv

# macOS/Linux:
source .venv/bin/activate

# Windows Command Prompt:
.venv\Scripts\activate.bat

# Windows PowerShell:
.venv\Scripts\Activate.ps1

pip install -r requirements.txt

2. Export your API key as an environment variable:

export ANTHROPIC_API_KEY="sk-ant-..." # Windows cmd: set ANTHROPIC_API_KEY="sk-ant-..." # Windows PowerShell: $env:ANTHROPIC_API_KEY="sk-ant-..."

3. Run the script:

python basic_agent.py

Output

When you run the script, the agent reads the files it is allowed to read and streams the messages of the loop to the terminal, ending with its summary.