WikipediaLoader endup in JSONDecodeError

LangChain Academy
,
1

In Module 5 research-assisant, (or in anyplace that uses wikipedialoader), my invoke stage akways end with ‘Wikipedia failed: JSONDecodeError(‘Expecting value: line 1 column 1 (char 0)’)’ , direct calling with requests api works fine, seems to be some issue with the wikipedialoader api?

Not a big issue but it’s been bugging me a lot, wondering if anyone have the same issue.

Attaching trace from langsmith

image
image1201×1018 117 KB

2

hi @Learium

the WikipediaLoader from langchain-community delegates to the unmaintained goldsmith/Wikipedia package, which calls r.json() with no fallback. As of March 2026, Wikimedia opened goldsmith/Wikipedia#350 warning that the package’s hard-coded User-Agent (wikipedia (https://github.com/goldsmith/Wikipedia/)) would be severely rate-limited starting April 2026. When rate-limited, Wikimedia returns an empty/HTML body, so r.json() raises the exact Expecting value: line 1 column 1 (char 0) error. Direct requests works because it lands in a different rate-limit bucket.

Fixes (in order):

  1. One-line patch: wikipedia.set_user_agent("research-assistant/0.1 (you@example.com)") before importing WikipediaLoader.
  2. Wrap the call to swallow JSONDecodeError.
  3. Switch to the actively-maintained wikipedia-api package (forces UA in constructor).
  4. Use raw requests against the MediaWiki Action / REST API (the user already verified this works).
  5. Authenticate via Wikimedia OAuth for production-grade limits.

Also noted that langchain-community could plausibly accept a user_agent kwarg on WikipediaAPIWrapper and translate the cryptic JSONDecodeError into something actionable - a small contribution worth filing upstream.

3

Thanks! Fix one indeed worked but the request takes about 3-5mintues, is it normal behavior?

4

Hmmm, quite long… you mean pure http request takes so long?

5

Soz got busy with work. Pure http request is quick, the call using the provided library with one-line patch takes super long.