DEV Community

Mani
Mani

Posted on

The JSON Hiding Behind the Anthropic SDK

#python #learning #ai

I got the Claude API working in about ten minutes, just following Anthropic's Get Started guide and copying the code as I went. A few lines of Python, a prompt, a reply.

Done.

Then I printed the reply to see what I'd actually gotten back:

print(message.content)
Enter fullscreen mode Exit fullscreen mode 
[TextBlock(citations=None, text='Here are some...', type='text')]
Enter fullscreen mode Exit fullscreen mode

Why is it a list? I asked one question. I thought I'd just get an answer as plain text.

Here's what I eventually understood: client.messages.create() is a wrapper. The Messages API itself does one thing — you send it JSON, it sends JSON back. The Python method just builds that JSON, posts it over HTTP, and turns the reply back into Python objects.

So the call I wrote is really this request:

{
  "model": "claude-opus-4-8",
  "max_tokens": 1000,
  "messages": [
    { "role": "user", "content": "What should I search for..." }
  ]
}
Enter fullscreen mode Exit fullscreen mode

And the server hands back this response:

{
  "role": "assistant",
  "content": [
    { "type": "text", "text": "Here are some search strategies..." }
  ],
  "stop_reason": "end_turn",
  "usage": { "input_tokens": 25, "output_tokens": 530 }
}
Enter fullscreen mode Exit fullscreen mode

(The real response has a few more fields -- cache and token details -- but I've trimmed it down to the parts that matter here.)

Look at content on both sides. It's a list of typed blocks.

When I wrote "content": "What should I search for..." as a plain string in Python, that was just shorthand -- underneath, it's one text block in a list.

That's the whole answer to "why is it a list?" The response wasn't being weird. "content" is always a list of blocks, going out and coming back.

The string form just hides it on the way out.

Top comments (0)