Skip to main content

​
What You’ll Learn

  • Extracting clean content from one or many URLs
  • Basic vs advanced extraction depth
  • Query-focused extraction for targeted content retrieval
  • Batch extraction (up to 20 URLs in a single call)

​
How Does It Work?

Tavily Extract takes a URL (or list of URLs) and returns the page content as clean markdown or plain text. It handles JavaScript-rendered pages, removes boilerplate (ads, navigation, footers), and returns structured content ready for LLM consumption. Two extraction depths are available:
DepthSpeedSuccess RateContentCost
basicFastGoodStandard page content1 credit per 5 URLs
advancedSlowerHigherTables, embedded content, JS-rendered pages2 credits per 5 URLs

​
Getting Started

Get your Tavily API key

1

Install the Tavily Python SDK

uv venv
uv pip install tavily-python
2

Extract content from a URL

import os
from tavily import TavilyClient

client = TavilyClient(api_key=os.environ["TAVILY_API_KEY"])

response = client.extract(
    urls="https://en.wikipedia.org/wiki/Artificial_intelligence",
    extract_depth="advanced",
)

result = response["results"][0]
print(f"URL: {result['url']}")
print(f"Content length: {len(result['raw_content'])} chars")
print(result["raw_content"][:500])
3

Output

URL: https://en.wikipedia.org/wiki/Artificial_intelligence
Content length: 48231 chars
# Artificial intelligence

**Artificial intelligence (AI)**, in its broadest sense,
is intelligence exhibited by machines, particularly
computer systems. It is a field of research in computer
science that develops and studies methods and software
that enable machines to perceive their environment and
use learning and intelligence to take actions...

​
Batch Extraction

Extract content from up to 20 URLs in a single call. Failed URLs are reported separately without blocking successful ones. 
import os
from tavily import TavilyClient

client = TavilyClient(api_key=os.environ["TAVILY_API_KEY"])

urls = [
    "https://en.wikipedia.org/wiki/Artificial_intelligence",
    "https://en.wikipedia.org/wiki/Machine_learning",
    "https://en.wikipedia.org/wiki/Data_science",
]

response = client.extract(urls=urls, include_images=True)

for result in response["results"]:
    print(f"{result['url']}: {len(result['raw_content'])} chars")

if response["failed_results"]:
    for fail in response["failed_results"]:
        print(f"Failed: {fail['url']} - {fail['error']}")

​
Query-Focused Extraction

When you pass a query parameter, Extract reranks the content chunks by relevance to your question. Combined with chunks_per_source, this returns only the most relevant portions of each page. 
import os
from tavily import TavilyClient

client = TavilyClient(api_key=os.environ["TAVILY_API_KEY"])

response = client.extract(
    urls="https://en.wikipedia.org/wiki/Artificial_intelligence",
    query="What are the main ethical concerns with AI?",
    chunks_per_source=3,
)

print(response["results"][0]["raw_content"])
The raw_content field will contain the top 3 most relevant chunks separated by [...], rather than the full page content. This is useful for keeping LLM context windows small while maintaining relevance.

​
Choosing the Right Extraction Depth

  • Static HTML pages (blogs, articles, documentation)
  • When speed matters more than completeness
  • High-volume batch jobs where cost is a concern
  • Pages with straightforward content structure
  • JavaScript-rendered single-page applications
  • Pages with tables, charts, or embedded content
  • When you need the highest success rate
  • Complex pages where basic extraction misses content

​
Critical Knobs

  • "basic" (default) — standard HTML pages, 1 credit per 5 URLs
  • "advanced" — JS-rendered pages, tables, embedded content, 2 credits per 5 URLs
  • Pass a query to rerank content by relevance to your question
  • Pair with chunks_per_source (1–5) to return only the top snippets
  • Without query, full page content is returned
  • "markdown" (default) — preserves headings, links, and structure
  • "text" — plain text, lighter for simple pipelines
For the complete parameter list, see the Extract API reference.

​
Next Steps

Extract API Reference

Full parameter list, response schema, and interactive playground.

Extract Best Practices

Depth selection, two-step search-then-extract, and optimization tips.

Python SDK Reference

Python client methods, async support, and type details.

JavaScript SDK Reference

JavaScript/TypeScript client methods and usage.