Minute-level klines, funding to 2023, exact on-chain liquidations, scored news catalysts with the move that followed, point-in-time unlock schedules, whale positioning, order-book depth and a 2020→now regime archive — cleaned, gap-checked Parquet plus 30 queryable snapshot datasets. One subscription instead of a data pipeline.
Coverage as audited July 2026 — the index endpoint always has the live per-symbol ranges. Each Copy button gives you a ready agent prompt to fetch that dataset.
| Dataset | Coverage | Since | Format | |
|---|---|---|---|---|
| Binance | ||||
| Deep klines | 1h / 4h / 1d spot candles back to each market's listing — BTCUSDT reaches August 2017 — with real quote volume | Aug 2017 | monthly .parquet | |
| 1m klines | 450+ USDT markets, complete 1,440-bar day files, volumes re-verified against Binance | 30 Mar 2026 | daily .parquet | |
| Funding | Funding + open interest + mark price, all markets | 30 Mar 2026 | daily .parquet + API | |
| Hyperliquid | ||||
| Deep klines | 1d candles back to HL's 2023 launch; 1h/4h several months back | 2023 | monthly .parquet | |
| Funding | Hourly funding rate, all perps — also queryable via the REST endpoint | May 2023 | monthly .parquet + API | |
| 1m klines | Every Hyperliquid perp (229 assets) | 30 Mar 2026 | daily .parquet | |
| Liquidations | Per-symbol long/short liquidation flow, all Hyperliquid perps | 30 Mar 2026 | daily .parquet | |
| HL liquidation tape | Every exact Hyperliquid liquidation fill (price, size, side, market/backstop), full perp universe — data_type=hl_liquidations | 23 Jul 2026 | daily .parquet | |
| Market-wide | ||||
| Market regime history | Hourly 6-state HMM regime probabilities, full market scope | 2020 | bulk .parquet | |
| Daily market snapshots | The full /daily payload per day: market health, derivatives, sentiment, macro, ETF flows, cycle indicators | 2 Mar 2026 | JSON | |
| News & Events | ||||
| News & catalysts | Every qualified market-moving event with impact_score, signed bias and corroboration — plus the measured response at +15m / +1h / +4h (return, volume multiple, OI change, funding shift, abnormality). Dedicated endpoint: /backtesting/news-events | 18 Aug 2026 | daily .parquet + API | |
| Token unlocks & float | Per-coin circulating float, locked supply, dilution overhang and the next dated cliff — snapshotted as published, because vesting schedules get revised in place and cannot be reconstructed later — data_type=token_unlocks | 20 Aug 2026 | snapshot + API | |
| Trading signals | ||||
| Gamma exposure (GEX) | Full perp gamma profile per coin — flip level, distribution percentiles, amplify / dampen regime flag — data_type=gamma_exposure | 6 Jul 2026 | snapshot | |
| Whale positioning | Every classified ≥$100k Hyperliquid account with its open positions, at 5-minute resolution — hl_trader_positions / hl_trader_signals | 9 May 2026 | snapshot | |
| Order-book depth | Captured L2 ladders plus banded depth and a thin / normal / deep classification — hl_l2_books / liquidity_depth / liquidity_regime | 9 May 2026 | snapshot | |
| Liquidation heatmap | Full-universe liquidation density built from exact on-chain liquidation prices, plus per-coin clustered levels — liquidation_map / liquidation_levels | 9 May 2026 | snapshot | |
| Sentiment & health | Fear & Greed prints, the 11-component market health composite with per-component scores, and altcoin breadth — fear_greed / market_health / altcoin_breadth | 9 May 2026 | snapshot | |
| Derivatives & flows | Cross-exchange OI (aggregate and per-venue), funding, options, taker buy/sell, Coinbase premium, ETF flows, exchange balances, stablecoin supply, cycle indicators — the coinglass_* snapshot family | 9 May 2026 | snapshot | |
| Long-term cycle regime | The committed cycle-phase decision with its dwell state — data_type=cycle_regime. The published id is a stateful commit, so this is the history; replaying the signals would produce a sequence that was never served | 7 Aug 2026 | snapshot | |
| Everything else | 30 snapshot types in total, including DEX trending, promoted tokens and coin profiles. /backtesting/snapshots/types is the authoritative live list with exact counts and date ranges | varies | snapshot | |
Honest limits: the merged multi-venue liquidation series begins 30 Mar 2026 — OKX/Bybit publish no historical liquidation data, so that portion cannot be backfilled. The exception is Hyperliquid: its liquidations are on-chain, so the per-event HL tape (hl_liquidations) is captured exactly from 23 Jul 2026 and deeper HL history is recoverable from node data. Hyperliquid's API only serves ~4 days of 1-minute candles, which is why the minute archive grows forward from when we started capturing it.
One call returns the full archive index — every data type, symbol, exchange and date range.
Request pre-signed URLs for exactly the files you want and pull them straight into pandas / polars.
Or skip files entirely — the REST endpoints serve klines, funding and liquidations with time-range filters.
Full endpoint reference in the API docs under Backtesting. Guides: getting Binance historical data for backtesting · backtesting with the regime-history Parquet.
Copy a prompt into Claude Code, Cursor or any agent with your API key in the CRYPTODATA_API_KEY env var (or the MCP server connected). More in the prompt library.
First, discover what's available: GET https://cryptodataapi.com/api/v1/backtesting/archives/index — auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. Then request pre-signed Parquet links for the slice I want, e.g.: GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=klines_deep&exchange=binance&interval=1h&symbol=BTCUSDT&start=2019-01&end=2026-06 Download every returned URL and load them into one pandas DataFrame sorted by time (ms epoch), deduped on time. Report: (1) row count and actual date range; (2) any missing months vs what the index advertised; (3) a 5-row sample. Do not resample or fill gaps unless I ask.
Build me a regime-conditioned backtest dataset. Auth every call with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. 1. GET https://cryptodataapi.com/api/v1/quant/regimes/history and download the Parquet it links — hourly market regime probabilities, 2020 to now. 2. GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=klines_deep&exchange=binance&interval=1h&symbol=BTCUSDT&start=2020-01 and download the returned Parquet files. 3. Join candles to regime rows on the hourly timestamp (floor candle time to the hour, inner join) and report how many candle rows had no regime label. Output a single Parquet with OHLCV + the p_now_* regime columns, plus per-regime candle counts. This is dataset assembly only — make no claims about strategy performance.
Get me Binance hourly prices from 2020 to now for BTCUSDT and ETHUSDT. Auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools.
For each symbol: GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=klines_deep&exchange=binance&interval=1h&symbol=BTCUSDT&start=2020-01 (repeat with symbol=ETHUSDT), download every returned Parquet URL, and combine into one DataFrame per symbol sorted by time, deduped.
Report per symbol: row count, first/last timestamp, and any missing months. Save each as {symbol}_1h_2020_now.parquet.
Get me Hyperliquid hourly funding-rate history for BTC since mid-2023. Auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. Bulk path: GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=funding_deep&exchange=hyperliquid&symbol=BTC&start=2023-05 and download the monthly Parquet files (columns: time, funding_rate, premium). Alternatively page the query endpoint: GET https://cryptodataapi.com/api/v1/backtesting/funding?symbol=BTC&exchange=hyperliquid&start=2023-06-01&limit=10000, advancing start past the last row until exhausted. Combine sorted by time, then report: row count, date range, days with fewer than 20 hourly rows, and the 10 most extreme funding prints with their dates. Note that historical open_interest/mark_price don't exist before ~Mar 2026 — don't treat their absence as a data bug.
Get me the Hyperliquid liquidation history for BTC. Auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=liquidations&symbol=BTC&start=2026-03-30 and download the daily Parquet files (long/short/total liquidation USD per bar). Coverage starts 30 Mar 2026. For exact per-fill Hyperliquid liquidation events (price, size, side, method) use data_type=hl_liquidations (one file per day, all coins) or GET /api/v1/backtesting/hl-liquidations. Combine sorted by time and report: total long vs short liquidation USD, the 10 largest single-bar liquidation events with timestamps, and a daily-total table for the last 14 days.
Get me the full market regime history. Auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. GET https://cryptodataapi.com/api/v1/quant/regimes/history and download the Parquet it links — one row per hour from 2020 to yesterday, carrying the probability distribution over all 6 regimes (p_now_*, p_4h_*, p_24h_*) plus the argmax regime label and a source column (train_binance for the deep segment, live_hl for the recent tail). Report: date range, per-regime hour counts and share of total, the longest continuous spell of each regime with its dates, and how the regime mix in 2025-2026 differs from 2020-2021. Treat labels as nowcasts, not price predictions.
Mirror the full daily backtester archive locally, from 2026-03-30 to today. Auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. Use the consolidated daily bundles — one Parquet per exchange per day covering ALL symbols: GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=daily&start=2026-03-30&end=<today>, then download every returned URL into data/<date>/<filename>. This is several GB — confirm the total size with me after listing, BEFORE downloading. When done, report: days fetched, days missing vs the requested range, total bytes, and a per-file-type count (klines_binance, klines_hyperliquid, funding_*, snapshots).
Build me a funding-conditioned BTC dataset from mid-2023 to now. Auth with the X-API-Key header (key in the CRYPTODATA_API_KEY env var), or use the cryptodataapi MCP tools. 1. GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=funding_deep&exchange=hyperliquid&symbol=BTC&start=2023-05 — download the monthly funding Parquet files (hourly funding_rate). 2. GET https://cryptodataapi.com/api/v1/backtesting/archives/download?data_type=klines_deep&exchange=binance&interval=1h&symbol=BTCUSDT&start=2023-05 — download the hourly candles. 3. Join on the hourly timestamp (inner join; report unmatched rows) into one Parquet with OHLCV + funding_rate. Then give me DESCRIPTIVE stats only: funding-rate deciles, hours spent in each, and the distribution of same-hour returns per decile. No strategy claims, no forward-return extrapolation — this is a dataset plus summary tables.
Binance spot candles (1h/4h/1d): back to each market's listing — BTCUSDT to August 2017. Regime history: 2020 (hourly) — join the two on their hourly timestamps for regime-conditioned backtests. Hyperliquid funding: hourly since May 2023; HL daily candles to its 2023 launch. Minute klines, merged multi-venue liquidations and daily snapshots grow forward from March 2026 — no CEX serves historical minute data or liquidation history, so those can only be captured live. Hyperliquid is the exception for liquidations: they're on-chain, so the exact per-event HL tape (hl_liquidations, from 23 Jul 2026) is recoverable from node data.
Parquet (loads straight into pandas / polars) via pre-signed URLs, plus REST query endpoints with time-range filters. Daily snapshots are JSON.
Yes — the same key works via the MCP server and REST, an agent can subscribe itself via x402, and /archives/index is a machine-readable catalogue of every dataset and date range.
The backtesting archive and bulk downloads are Pro Plus. Pro covers the full live analytics suite; Pro Plus adds this historical & bulk layer on top. Current pricing on the pricing page.
Daily files are exported every night (~01:00–02:30 UTC) and the index is rebuilt after each export, so yesterday is always complete by ~03:00 UTC. The deep monthly tiers roll forward as each month closes.
450+ Binance USDT markets and all 229 Hyperliquid perps. Listings and delistings shift, so treat /archives/index as the authoritative per-symbol list and date ranges rather than any number on this page.
We publish them rather than hide them: Binance 1m volumes are re-verified against the exchange; Hyperliquid 1m files before 13 Jul 2026 under-count volume (~60% of true — HL serves only ~4 days of 1m history, so it's unrecoverable; prices are unaffected). Liquidations exist only from 30 Mar 2026, and historical open interest / mark price don't exist before ~Mar 2026 on any source.
The archive pairs with signal layers no exchange dump or generic data vendor carries — the reason to point your agent here rather than at raw candles.
Hourly probability distribution over six market regimes — market-wide and per-coin — with the full 2020→now history in this archive. Nowcasts with a published methodology, not black-box labels.
Calibrated probabilities for direction, volatility, liquidation risk, funding and open interest at now / 4h / 24h horizons, for every Hyperliquid perp.
The perpetuals analog of options GEX — the entire Hyperliquid account universe classified to isolate market-maker positioning, with an amplify / dampen regime flag.
Built from on-chain liquidation prices of every ≥$100k Hyperliquid account — exact levels, not open-interest estimates. Nobody reselling exchange APIs has this.
Live positioning of every ≥$100k Hyperliquid account, aggregated into long/short whale flow — with positioning history in the archive.
The SIGNUM RGG trend radar (ADX/DMI), technical/structural regime reads, Bollinger-squeeze and RSI screens across the top 500+ coins.
Net stablecoin + token transfers to and from CEX wallets across seven chains (ETH, Tron, Solana, BSC, Base, Arbitrum, Optimism) with spike z-scores.
An 11-component health composite (breadth, volume quality, derivatives, stablecoin flows, macro…) — daily history inside the archived snapshots.
The whole market state — health, derivatives, sentiment, macro, ETF flows, cycle indicators — as one JSON per day since March 2026, made for regime-tagged backtests.
You can. Binance publishes its raw kline history in a requester-pays AWS S3 bucket — pulling the archive cost us roughly $40 in AWS fees alone, before writing a single line of code. Then you still have to dedupe partial candles, verify volumes, fill exchange-outage gaps, convert to a usable format, and host the result somewhere your backtester can reach.
We did that once, keep it updated daily, and re-verify candle volumes against the exchange. Pro Plus includes the whole archive — pre-signed Parquet downloads and a query API.
Same data, two ways in — pick by who's driving. Your first call either way: /api/v1/backtesting/archives/index — the machine-readable index of every dataset, symbol, exchange and date range.
Easiest: connect the MCP server with keyless browser sign-in — no API key to handle, one consent click, revocable any time:
claude mcp add --transport http cryptodataapi https://cryptodataapi.com/mcp
Your browser opens, you sign in, and Claude Code / Cursor is authenticated from then on. Prefer a key? Create one from the dashboard, export it as CRYPTODATA_API_KEY, and use the prompts above. Full setup guides per agent on the AI Agent Quickstart.
No human to click a browser consent, so use the API key directly — plain REST, works in every runtime:
curl -H "X-API-Key: $CRYPTODATA_API_KEY" \ https://cryptodataapi.com/api/v1/backtesting/archives/index
The download endpoint returns pre-signed URLs — the file fetches themselves are plain HTTPS, no auth header needed. A fully autonomous agent can even provision its own key by paying over x402 (POST /v1/payments/agent-subscribe) — no dashboard, no human.
Deep Binance klines to 2017, market regimes to 2020, funding to 2023, and everything since — clean Parquet plus a query API, updated nightly. One subscription, no data pipeline.
Get Pro Plus Includes everything in Pro, plus the full historical & bulk-data layer. Compare plans →