How does an AI agent detect crypto chart patterns like double tops and head & shoulders?

Chart patterns vs candlestick patterns

Two different shapes of pattern, both returned in the same analysis.patterns array:

Chart patterns describe the macro shape of price over a stretch of the series; that is why each one reports a span — the number of candles the formation covers. Candlestick patterns are local, single- or two-bar signals. The brief returns both classes in one call, so your agent never has to run two passes. If you want detail on the 1–2 bar signals, see the crypto candlestick patterns API.

The one call

POST https://patternfetch.com/v1/brief
Authorization: Bearer pf_…
Content-Type: application/json

{ "ticker": "ETH/USDT",
  "timeframe": "1d" }

"patterns": [
  { "name":"head_and_shoulders",
    "confidence":0.74, "at":177, "span":40 },
  { "name":"double_bottom",
    "confidence":0.86, "at":158, "span":14 }
]

The same call also returns support/resistance levels, the regime (trend + volatility), interpreted indicators (RSI/EMA), and a one-line nl summary your agent can act on — all in a few hundred tokens. See it live →

How confidence + span work

Each detected pattern carries three fields that an agent can reason over directly:

So { "name":"head_and_shoulders", "confidence":0.74, "at":177, "span":40 } reads as: a head-and-shoulders that completes at candle 177 and covers the preceding 40 bars, with a clean-but-not-textbook geometric fit of 0.74.

Free to start

For AI agents (MCP)

Call the tool patternfetch_brief over MCP (Streamable HTTP) and read analysis.patterns for the chart patterns. Two ways to connect, both zero-friction:

A) OAuth — one-click connect (no key handling)

Add the server URL and click Authorize once — patternfetch mints a free-tier key for you automatically, nothing to paste.

# Claude.ai → Settings → Connectors → Add custom connector
https://patternfetch.com/mcp
# (discovery: /.well-known/oauth-protected-resource)

B) Bearer key (config file)

{
  "mcpServers": {
    "patternfetch": {
      "url": "https://patternfetch.com/mcp",
      "headers": { "Authorization": "Bearer pf_…" }
    }
  }
}

Or one line: claude mcp add --transport http patternfetch https://patternfetch.com/mcp --header "Authorization: Bearer pf_…"

The honesty layer: every pattern ships its base rate

Most pattern and signal APIs hand a model a confident-looking "double_bottom", 0.86 and let it quote that blind. patternfetch attaches an evidence block to every detected pattern — the backtested historical hit-rate for that exact pattern, timeframe and confidence band, the pattern-free baseline for the same market and horizon, and a cluster-robust interval on the difference — measured with no lookahead (chart patterns are scored only from the bar they become knowable). A hit rate on its own cannot be read: a drifting market moves it away from 50% with no pattern involved. Measured against the baseline instead, three of the 105 buckets we tested clear an uncorrected cluster-robust 95% interval — and all three are crypto: bullish_engulfing 1h −1.76pp (±1.36, n=8,897), double_bottom 1d −12.25pp (±8.91, n=143), and double_top 1d +9.78pp (±8.42, n=149), the only bucket of the 105 that lands above its baseline. That is fewer than the ~5.3 chance alone would produce at this threshold, and none of the three survives a Bonferroni correction across 105 comparisons; on US stocks & ETFs it is 0 of 60. The evidence block reports exactly that, instead of laundering a shape score into a confident call. Impersonal market data, not advice. See the method →

FAQ

Related