Web search is for text. Subsets is for numbers.

Search and SQL over up‑to‑date datasets from official statistical sources, cleaned into plain relational tables. Every number traces back to the open-source run that produced it.

# Claude.ai → Settings → Connectors → Add custom connector
https://mcp.subsets.io

Remote tools for Claude.ai, Claude Code, Cursor, and Codex CLI. Signs you in on first use. See docs →

Web search vs. Subsets

1 / 4

Point lookup

Ask both: “How high did Australian inflation spike during the Korean War wool boom?”

Web searchvague

“Inflation surged in the early 1950s — retrospectives suggest prices rose by somewhere around 20% at the peak of the boom.”

Paraphrased from a retrospective article · no primary source, no exact figure

With Subsets

SELECT obs_date, value_text
FROM "reserve-bank-of-australia-g1-data"
WHERE series_id = 'GCPIAGYP'
  AND obs_date BETWEEN '1951-01-01' AND '1951-12-31'

Q4 1951 · +25.6% year on year

Source

Reserve Bank of Australia · G1 — Consumer price inflation · view the dataset →

Where every number comes from

Official source

rba.gov.au · raw release

Statistical agencies and central banks like the ONS, RBA, and Bundesbank publish the raw data.

Open-source connector

subsetsio/connectors

The exact ingestion code is public. Read it, fork it, file an issue.

Public CI run

public run logs

Every table is built in a logged workflow run, one click from its dataset page.

Versioned table

rba g1-data · pinned

Revisions are tracked and reviewed. Your agent queries a pinned version.

Read the connector code →View connector status →