AI-detected chart patterns on a coin, with bullish/bearish signal and confidence. One call returns a multi-timeframe set — no list endpoint, no pagination.
20 patterns per coin per request, across 1D, 1M, 3M, 1Y, 3Y and ALL windows. There is no list endpoint, and the response is a bare JSON array — not an envelope.
Every field the corpus returns for one record. Real shape — walked corpus-wide, not sampled.
A curl that runs against the live API right now, plus the MCP tool names for the same data.
MCP tools
get_kasandra_patternsThree concrete questions this capability answers that a general crypto API does not.
signal is bullish or bearish, confidence is in [0, 1]. A 'high-confidence bullish, recent' screen is a one-liner.key_points: [{price, timestamp}] with the trigger prices and the timestamps in epoch milliseconds.Required, not optional. Coverage gaps, staleness and parameter traps that were observed by running the curl above against the live server on 2026-09-18.
Keyed on coin slug, not ticker.
/kasandra/patterns/bitcoin/ returns 20 patterns; /kasandra/patterns/BTC/ returns 404 with detail "coin requested does not exist or is not supported". The slug is the join key, the same convention the news / market endpoints use, but a reader used to a ticker-keyed price feed will type the ticker first and the page must call this out.
There is no list endpoint.
/kasandra/patterns/ returns 404. There is no corpus-wide count of patterns and no list-of-coins endpoint; the coin you can query is whatever the server happens to support, discoverable only by trying the slug.
The response is a bare array of 20, not an envelope.
There is no {count, next, results} shape. Each call returns at most 20 patterns for one coin, and the set covers a fixed six-interval span (1D, 1M, 3M, 1Y, 3Y, ALL). Iterate interval rather than paginating.