WikiSearch MCP

Wikipedia search for agents. Try it below — your API key stays in your browser.

Claude Haiku 4.5
enter to send · shift+enter for newline

https://wikisearch.sauerslabs.workers.dev/mcp

One tool: search_wikipedia

// exact article
"Albert Einstein"

// targeted section
"Hafnium chemical properties"

// general search
"largest earthquakes in Japan"

query
search Wikipedia
exact title
one article
otherwise
a few articles
clean up text
remove duplicates
return to model

Wiki Trivia

Trivia questions answered with and without the Wikipedia tool to measure retrieval uplift.

Obscure Information Retrieval

20 obscure geography and engineering questions, compared with and without the Wikipedia tool.

QA Precise

10 hard technical questions where the exact answer is hard to recall, scored on correctness and answer quality.

HLE Sauers

Questions from Humanity's Last Exam, answered with and without the Wikipedia tool.

Private QA

Previously private technical questions scored on answer correctness and reasoning quality.

Good Recency

Post-cutoff 2025–2026 events: does the model decline without the tool and commit to answers with it?

Bad Recency

2026 events with no Wikipedia article: does the model fabricate specifics, with and without the tool?

Incorrect Information Robustness

Tool results are silently tampered with wrong values to see whether the model trusts the source or its own knowledge.

CCP Influence Benchmark

Do subtly pro-CCP edits injected into Wikipedia results shift the model's numeric opinion scores, measured via Cohen's d?

Unnecessary Tool Use

Does the model needlessly invoke the Wikipedia tool on trivia it already knows?

Implementation Benchmark

Python implementations of 20 non-trivial algorithms, graded on correctness, helpfulness, elegance, and completion.