Polymarket Data Access

• Public exchange: markets, current books, activity — free, keyless, from the official API.
• Starter archive: 3 days of 250ms history per market, CSV samples, and the documentation — no card.
• Pro archive: 30–120 days of crypto, 30 of sports, more depth and rate headroom, for research volume.
• Enterprise: bulk Parquet/CSV/JSON to your S3 bucket, partitioned by day and market type — for lakes.

• Apps and dashboards read the official live feed; the free tier covers prototypes and watchlists.
• Researchers buy Starter to validate, then Pro for the record; both tiers keep the same schema and endpoints.
• Data teams land the bulk delivery into their lake; the REST API serves the live slice their apps consume.
• AI agents reach the same endpoints through the MCP server, so assistant workflows share the identical response shapes.

The free tier is deliberately useful enough to validate against reality — three days of 250ms frames is a complete small study — while paid tiers buy depth, not access to anything exotic.

Every tier shares field names and timestamps, so upgrading comfortable is the design; nothing changes shape when you move up the ladder.

• Write a single client that can talk to the official public API and the archive based on a config flag; run it against both for the same market.
• Expose the source and timestamp on every response; the provenance column is the audit trail that makes the access ladder verifiable.
• Take one market through all four rungs — public, Starter, Pro, bulk — and note what changed at each step: schema stayed, depth and window grew.
• Document your tier choice in the repo: the decision record says why three days was enough or why it was not, so the next researcher skips the debate.

The access model is not a moat but a ladder, and every rung speaks the same field names — the deliberate design that makes upgrading a config change rather than a migration.

The measured value of the paid tiers is depth, not exclusivity: the free record is genuinely sufficient for validation, and Pro is for when the study needs the full window.

Know the limits of each rung before you claim them: Starter's 3 days, Pro's windows, bulk's partitioned Parquet — the page is the honest map of what each affords.

Apps read the live public feed; research reads the archive; lakes read the S3 delivery — three consumers, one schema, and provenance recorded at every hop.

The MCP server extends the same response shapes to AI agents, so the access ladder serves humans and assistants identically.

Data access on Polymarket is a ladder with public, free, paid, and bulk rungs — each real, each useful, and each honest about what it covers.

Nothing on a higher rung is a different format; everything is the same schema with more depth and history.

Choose the rung by the requirement of the study, and keep the free tier in your toolkit forever for rapid validation.

A useful decision aid for choosing a rung: write the study's data requirement as a sentence — "I need N markets, M days, 250ms frames" — and map it to the ladder: free covers prototypes, Starter three days, Pro the window, bulk the lake.

When in doubt, run the study twice, once on the free tier and once on paid; if the free version answers the question, the paid spend was optional, and if it does not, the spend is justified by the answer.

The page's conclusion is that access is a ladder with honest prices at every rung, and the same schema at every height — the choice is a research decision, not a data-quality one.

• data download
• mcp server
• data with s3