To read a known web page with Jina Reader in Python, request https://r.jina.ai/ followed by the full source URL. Keep the returned text with its source address, then check that the passages your application needs survived extraction. This tutorial covers one URL at a time; it does not implement site discovery or browser actions.
Request a reviewed public URL
Install requests with python -m pip install requests. The example accepts an optional JINA_API_KEY environment variable. Check the current official account limits before choosing a worker count. A timeout, non-success HTTP status, or empty body is a failed extraction, not a valid document.
import os
from urllib.parse import urlsplit
import requests
def read_jina(url, session=requests):
if urlsplit(url).scheme != "https":
raise ValueError("Use a reviewed public HTTPS URL")
headers = {"Accept":"text/plain"}
if os.environ.get("JINA_API_KEY"):
headers["Authorization"] = "Bearer " + os.environ["JINA_API_KEY"]
response = session.get("https://r.jina.ai/" + url, headers=headers, timeout=60)
response.raise_for_status()
if not response.text.strip():
raise ValueError("Empty Jina response")
return {"source_url":url, "text":response.text}
if __name__ == "__main__":
result = read_jina("https://www.searchcans.com/reader-api/")
print(result["source_url"], len(result["text"]))
Validate the output before giving it to an LLM
Choose literal checks from the source: the main section heading, a table row including its unit, and any code block or link needed to answer the task. Record which checks failed. A response that contains readable prose can still omit the exact table the user asked about. Keep a source timestamp alongside the text and revalidate it when freshness matters.
For a retry, log the HTTP status and attempt count first. Do not silently loop on an authorization failure or assume a successful second call returned the same version. A bounded retry policy belongs in the application; this small example deliberately leaves retries visible to the caller.
Keep reading, discovery and selection separate
A Jina read request starts with a URL. If the application only knows a question, discover candidate URLs first and review their relevance and permitted use. For a SearchCans search response, the result list is data.organic and each URL is in link; for its Reader response, text is in data.markdown. Do not reuse a parser written for Jina plain text as a JSON parser.
For vendor selection, see Jina Reader alternatives by task. For billing units and workload calculations, see Jina Reader and SearchCans pricing comparison. The broader URL-to-Markdown Python tutorial covers the SearchCans request contract.