Marketlens

Recorded Polymarket order books across every market type, replayable to the tick.

Last updated

US persons
Yes
Taker fee
None
Settlement
Crypto
Liquidity
CLOB

What it is

A recorder for the part of Polymarket that Polymarket does not keep. The venue serves its current order book and a price series; it does not retain what the book looked like at a past moment. Marketlens subscribes to Polymarket's book stream, stores it, and sells it back as history: a full L2 snapshot about every 60 seconds, every price_change delta between snapshots at millisecond resolution, every executed trade with price, size and side, and OHLCV candles built from trades at resolutions from one second to one day. The claimed construction is that snapshot n+1 equals snapshot n plus the deltas between them, so the book at any tick can be rebuilt exactly.

What sets it apart from the other recorders in this category is breadth rather than depth. It tracks every recurring Polymarket family — crypto up-or-down from five-minute to daily, per-game sports and esports books, daily temperature buckets by city, quarterly earnings markets, macro and index strikes — and since 15 August 2026 admits any standalone event that sustains 10,000 USD of 24-hour volume. Its own API page, read on 30 September 2026, counted 1,054,008 markets, 1,033,591 of them resolved with the winning outcome stored, 43.0 billion order-book rows and 294 million trades.

The second half of the product is a Python backtester that replays those books with FIFO queue priority, configurable latency and slippage taken from recorded depth, in an order-level mode and a bar-level mode for slower signals. That is a capability of the data product, not a separate one: you hire it for the archive, and the replay engine is what makes the archive usable.

Availability

Anyone 18 or older who is not in a sanctioned country or on a denied-party list, per the terms of service. There is no identity check beyond an account, and no US restriction: the product is data about a venue, and reading it is not trading on it. One account per person — the terms allow suspending extra accounts opened to stretch the free allowance.

The API wants a bearer key on every request. An unauthenticated call to the markets endpoint answered 401 with UNAUTHORIZED on 30 September 2026, against a 404 for a fabricated path, so the service is live and gated rather than absent.

Pricing

Three tiers on the same endpoints, differing in how far back you may read, how many rows you may pull and how fast. Pro is 39 USD a month or 348 USD a year; Scale is 199 USD a month or 1,788 USD a year, the annual price being advertised as three months free. The free tier is not a trial: it is two million rows once, for markets open in the last seven days.

The unit that runs out is rows, not requests. A Parquet export charges its row count the first time you unlock a market and nothing on later downloads of the same file, and a series export can be priced with a dry run before anything is spent. On paid tiers, unused rows roll over for one month. Hitting the monthly row wall returns an error rather than a bill, and the SDK does not retry it.

Markets & resolution

Nothing here resolves anything. Marketlens stores the outcome Polymarket reached, and Polymarket's outcome is decided through the UMA optimistic oracle — so a stored winner is a copy of that process's result, including any market it resolved 50/50.

Coverage has a start date and gaps. Order-book history begins on 1 March 2026 and is continuous from 13 April 2026; the vendor's own FAQ lists "a few gap days" in March and April. The earliest candles are from 28 February 2026. Each market carries data_start and data_end fields for the window it was live in the archive, and a request outside that window returns DATA_NOT_AVAILABLE with the coverage attached.

Integrations

  • REST at api.marketlens.trade/v1, bearer-authenticated, cursor-paginated. Cursors expire after 24 hours.
  • Python SDK marketlens, MIT-licensed, Python 3.10 or newer, sync and async, returning pandas DataFrames. The first release, 0.1.0, went to PyPI on 2 March 2026, and on 29 September 2026 the repository had been pushed to that day and was not archived. The SDK retries 429s and 5xx with backoff, and deliberately does not retry the budget errors, which reset at fixed times.
  • Parquet exports per market or per series, served as one-hour presigned URLs, with a compact trade-aligned variant.
  • MCP server as an optional extra (marketlens[mcp]), running locally over stdio with your key: market search, point-in-time books, trades, candles, surfaces, and backtest tools that execute agent-written strategy code on your machine. An environment variable switches the backtest tools off.

There is no streaming feed for customers. For the live book, go to the venue's own CLOB API.

Limitations

  • Polymarket only. No Kalshi, no Limitless, no other venue.
  • Nothing before March 2026. The archive is a recording; a market that traded in 2025 is not in it at any price.
  • Snapshots are about a minute apart. Exact reconstruction between them depends on every delta having arrived. The terms say plainly that data may contain gaps from WebSocket disconnections or upstream outages, and no service-level agreement applies.
  • Redistribution needs a separate agreement. Paid tiers may use the data internally, derive from it and display it with attribution; exposing the raw data through your own API or reselling it is barred on every tier. Free-tier data is personal, non-commercial and research only. The terms also bar building "derivative financial products" on the data without written consent.
  • One person operates it. The operator named in the terms is an individual in Poland, not a company. That is not a defect in the data, but it is the continuity risk to weigh before building a pipeline on a proprietary archive you cannot re-record yourself.
  • The backtester is a model. Queue position, latency and slippage are simulated against recorded depth; the fill you would actually have received is not in the archive, because nobody placed it.

Alternatives

PolyOrderbooks serves Polymarket books in one-second buckets, for crypto markets only, priced by lookback window. DepthFeed covers crypto up-or-down markets on Polymarket, Kalshi and Limitless and has a WebSocket. Predexon sells tick history by the gigabyte across several venues, including Kalshi. If trades and prices are enough and the book is not, Polymarket's own Data API is free and reaches back to the start of every market at coarse resolution.

Specs

Interfaces
API, Python, MCP server
Export
Parquet, JSON
Available in
Global
KYC required
No
Market subjects
Crypto, Sports, Politics, Business, Macro, Culture
Resolved by
Resolves nothing
Maker fee
None
Platforms
Web, Library
AI features
None
Capabilities
Order book, Backtesting
Pricing verified
Availability verified
Markets verified
Capabilities verified

Background

How this part of the sector works, rather than which product to pick.

Also worth comparing

  • Goldsky Polymarket Datasets — Polymarket's on-chain fills and positions, backfilled and streamed into your own database.
  • Polymarket Data (SII dataset) — The "107 GB" open Parquet dump of Polymarket CLOB fills, November 2022 to March 2026.
  • Polymarket Data API — Polymarket's keyless read API for trades, positions, holders and price history.
  • DepthFeed — The recorded bid-ask ladder for crypto up-or-down markets, which no venue keeps itself.
  • PolyOrderbooks — Recorded L2 books for Polymarket crypto markets, priced by how far back you may look.
  • Predexon — Tick-level book history as Parquet, billed by the gigabyte, plus a mempool-aware feed.

FAQ

How far back does Marketlens order book history go?

To 1 March 2026, with continuous coverage from 13 April 2026 and a few gap days in March and April. Standalone events such as the 2028 election markets start on 15 August 2026. Nothing before March 2026 exists in the archive, because it was recorded live rather than reconstructed.

Does Marketlens cover Kalshi?

No. Every dataset, endpoint and catalogue page is about Polymarket. For a recorded Kalshi book, DepthFeed covers Kalshi crypto markets and Predexon sells Kalshi book snapshots by the gigabyte.

What does the free tier actually include?

Two million data rows, once — the allowance does not reset — against markets that were open in the last 7 days, at 180 requests a minute. Older markets answer with a history-window error. The terms limit free-tier data to personal, non-commercial and research use.

Can I publish charts or resell data from Marketlens?

Paid subscribers may display data and derived data in their own applications and reports, with attribution. Redistributing, reselling or exposing the raw data through your own API needs a separate written agreement on every tier, and free-tier data may not be used commercially at all.

Who runs Marketlens?

An individual. The terms of service, last updated 27 September 2026, name a private individual in Poland as the operator and state that the service is not run through a registered business entity. Disputes go to Polish courts under Polish law.