For people who build

QuizBase

A question API you can put in production without writing a content pipeline first.

1.1M+ questions, maintained

Go to QuizBase
QuizBase connected to Claude and other tools through an MCP server

Generating questions is easy. Keeping them alive is the work.

Every question is deduplicated against the rest of the pool, calibrated into five difficulty levels, attributed to the dataset it came from and versioned, so an edit six months from now does not silently change what your app served yesterday.

  • entityq 216,575
  • mintaka 154,325
  • kqa-pro 95,735
  • nq-open 86,714
  • opentriviaqa 48,862
  • creak 12,047
  • qasc 9,060
  • arc 7,787
  • mkqa 6,263
  • opentdb 5,149
  • webq 5,017
Approved English questions per dataset, after dedup - snapshot 12 May 2026

Three ways in

A REST API, a TypeScript SDK on npm and an MCP server, so an AI assistant can pull questions directly. Eleven open datasets underneath, English and Polish.

curl -H "X-API-Key: qb_pk_YOUR_KEY" \
  "https://quizbase.runriva.com/api/v1/questions/random?amount=5&lang=en"
QuizBase MCP server hooked into Claude Desktop - setup and first query

Measured, not promised

A load test on production recorded a 95th percentile of 101 ms at 50 requests per second, with no errors across 15,002 requests.

k6, production, 27 May 2026

  • discovery p50 · 22 ms p95 · 101 ms
  • browse p50 · 65 ms p95 · 143 ms
  • random p50 · 342 ms p95 · 571 ms
Response times in milliseconds - k6 against production, 27 May 2026, 15 002 requests, 0% errors

Free tier without a card

500 calls a day, the full production API, no card and no trial clock. Enough to build something real and find out whether it works.