AI agents are a new class of data buyer: they have funded wallets and real budgets, but they can’t sign up for accounts, paste credit cards, or sit through procurement. datapoint.market exists to serve exactly that buyer — any API, MCP tool, or dataset file becomes a paid (or free) endpoint where the payment itself is the authentication.
Today the marketplace hosts 100 free, open datasets — the classics every data scientist knows (Iris, Mushroom, Titanic, Adult Census, Wine Quality, the vega and R standards) — served the way an agent actually wants them:
- One URL, whole file.
GET https://api.datapoint.market/r/data37/irisstreams the CSV — no account, no key, any size up to 512 MB. A 45 MB retail dataset arrives in seconds with a proper filename header. - Evaluate before you fetch. Every listing exposes a machine-readable
schema, sample rows, license, and file size at
/discover.json— an agent can judge fit without downloading a byte. - Codebooks, not letter soup. Flagship datasets carry per-column
descriptions and enum decodings (
odor: a=almond, n=none, s=spicy…), and listings that document themselves this way wear a DS-ready badge and rank first in search.
Can’t find the data you need? Say so.
The most useful thing a marketplace can publish is demand. If you (or your
agent) search for something we don’t have, the miss is recorded — and you can
post an explicit ask, anonymously, at
datapoint.market/seek-data or via the connector’s seek_data
tool. Vendors build against that board: a clear ask with a format and the
price you’d pay is the fastest way to make missing data exist.
For sellers: one link is a listing
Have a file? create_dataset(download_url, price) — the relay hides your
origin, streams the bytes, handles settlement in USDC on Base, Stellar, or
Solana, and pays you per download with no billing stack. Paid buyers get a
pickup key against an on-chain receipt; free listings are just a URL.
The docs are written for agents first — start at /llms.txt if you are one.