Most crypto ML teams spend more time on their data pipeline than on their models.
Teams spend weeks on: connect to an exchange, normalize OHLCV, compute indicators, handle gaps, resample, repeat.
Before a single model trains, you've already burned engineering time on infrastructure that isn't your edge.
aarna team built the fix. Shipping "aTars MCP".
aTars is an MCP server that gives your agent or model direct access to a structured, pre-computed feature dataset across 9 major tokens: BTC, ETH, SOL, XRP and more.
What's in the dataset:
→ 40+ technical indicators already computed -- RSI, MACD, Bollinger, ADX, Ichimoku, OBV, VWAP and more
→ 1-min resolution OHLCV, 90-day rolling window (~129K candles per token)
→ ML-ready feature matrix with Pearson correlations and feature distributions built in
→ Resample to 1min / 1h / 4h / 1d depending on your model's frequency
→ Signal summaries pre-labeled BULLISH / BEARISH / NEUTRAL for supervised learning targets
You query it like a tool. Your model or agent gets a clean feature vector back. No pipeline. No normalization. No maintenance.
This is the data layer serious crypto ML work has been missing.
Free to access via MCP.
Endpoint in the replies.