How to beginner · 3 min read

How to extract JSON from Claude response

Q: How to extract JSON from Claude response

Use the anthropic Python SDK to send a prompt requesting JSON output from claude-3-5-sonnet-20241022. Parse the returned text with json.loads() to extract the JSON object safely from the response string.

Quick answer

Use the anthropic Python SDK to send a prompt requesting JSON output from claude-3-5-sonnet-20241022. Parse the returned text with json.loads() to extract the JSON object safely from the response string.

PREREQUISITES

Python 3.8+
Anthropic API key
pip install anthropic>=0.20

Setup

Install the official anthropic Python SDK and set your API key as an environment variable.

Run pip install anthropic>=0.20 to install the SDK.
Set your API key in your shell: export ANTHROPIC_API_KEY='your_api_key_here'.

bash

pip install anthropic>=0.20

Step by step

Send a prompt to claude-3-5-sonnet-20241022 asking for JSON output, then parse the response string to extract the JSON object.

python

import os
import json
import anthropic

client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])

prompt = (
    "Provide the following data as JSON only, no extra text:\n"
    "{\"name\": string, \"age\": integer, \"city\": string}\n"
    "Respond only with the JSON object."
)

response = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=200,
    system="You are a helpful assistant.",
    messages=[{"role": "user", "content": prompt}]
)

# The response content is a list with one element containing the text
text_response = response.content[0].text.strip()

# Parse JSON safely
try:
    data = json.loads(text_response)
    print("Extracted JSON:", data)
except json.JSONDecodeError as e:
    print("Failed to parse JSON:", e)
    print("Raw response:", text_response)

output

Extracted JSON: {'name': 'Alice', 'age': 30, 'city': 'New York'}

Common variations

Use async calls with asyncio and client.messages.acreate() for concurrency.
Adjust max_tokens and model for different Claude versions.
Use prompt engineering to enforce JSON-only output, e.g., "Respond only with JSON, no explanations."

python

import asyncio
import os
import json
import anthropic

async def main():
    client = anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
    prompt = (
        "Provide the following data as JSON only, no extra text:\n"
        "{\"name\": string, \"age\": integer, \"city\": string}\n"
        "Respond only with the JSON object."
    )

    response = await client.messages.acreate(
        model="claude-3-5-sonnet-20241022",
        max_tokens=200,
        system="You are a helpful assistant.",
        messages=[{"role": "user", "content": prompt}]
    )

    text_response = response.content[0].text.strip()

    try:
        data = json.loads(text_response)
        print("Extracted JSON:", data)
    except json.JSONDecodeError as e:
        print("Failed to parse JSON:", e)
        print("Raw response:", text_response)

asyncio.run(main())

output

Extracted JSON: {'name': 'Alice', 'age': 30, 'city': 'New York'}

Troubleshooting

If JSON parsing fails, check if the model output includes extra text or formatting; refine the prompt to enforce JSON-only output.
Use print(repr(text_response)) to debug hidden characters or whitespace.
Ensure your API key is set correctly in os.environ["ANTHROPIC_API_KEY"].

✅

Key Takeaways

Use the Anthropic SDK with claude-3-5-sonnet-20241022 to request JSON output explicitly.
Parse the response text with json.loads() to extract structured data safely.
Refine prompts to ensure the model returns JSON only, avoiding extra explanations or formatting.
Use async API calls for scalable, concurrent JSON extraction workflows.
Debug parsing issues by inspecting raw response strings and adjusting prompt instructions.

Verified 2026-04 · claude-3-5-sonnet-20241022

Verify ↗