wickworks

The dumb-as-rocks OHLC analyzer. You throw bars at it over HTTP, it throws indicators and SMC objects back. That's the whole product. No database, no queue, no state, no opinions, no "AI-powered signals," no upsell to a $97/mo Discord. Just primitives.

Every snake-oil-flavored TA SaaS out there wants to tell you when to buy. Wickworks tells you the order block is at 1.0832 and the RSI is 71.4. The "what does that mean?" part is where your strategy lives — and it should live in your code, not behind someone else's paywall.

Built on pandas_ta and smartmoneyconcepts, wrapped in a FastAPI server, locked behind 370 tests that diff our output against closed-form references on real EURUSD ticks. If a math bug slips in, the test suite screams before the container builds.

Table of Contents

• What's Inside
• Quick Start
• API
  • GET /health
  • GET /metadata
  • POST / — compute
  • The indicators object
  • Concepts
  • Available indicators
  • Request fields
  • Response
  • Errors
• MCP
• Agent integrations
• Configuration
• Architecture
• Development
• Testing philosophy
• License

What's Inside

All JSON field names are camelCase. Output is NaN-safe — NaN becomes null, never a literal NaN token that blows up downstream parsers. NumPy/Pandas scalars and arrays are serialized cleanly (no numpy.float64(...) leaks). Bars in UTC, math in UTC, container runs TZ=UTC — timezone bullshit is your problem, not ours.

Quick Start

docker run --rm -p 8000:8000 psyb0t/wickworks:latest

That's it. The service listens on :8000.

docker compose

services:
  wickworks:
    image: psyb0t/wickworks:latest
    ports: ["8000:8000"]
    environment:
      LOG_LEVEL: INFO
      MAX_BARS: "5000"
      MIN_BARS: "50"

Local development container

make run         # build the dev image and run uvicorn on :8000

API

Three endpoints. That's the whole surface.

GET /health

curl -s http://localhost:8000/health

{ "ok": true, "version": "0.7.0" }

GET /metadata

Returns the versioned catalog of labels, descriptions, units, categories, and interpretation notes for every output path. Consumers can cache it for the reported Wickworks version.

POST / — compute

Send OHLC(V) bars + the indicators you want. Get back only what you asked for — response keys mirror the keys you sent. No "let me also throw in 40 indicators you didn't ask for" energy.

curl -s -X POST http://localhost:8000/ \
  -H 'Content-Type: application/json' \
  -d '{
    "symbol": "EURUSD",
    "timeframe": "H1",
    "bars": [
      { "time": 1700000000, "open": 1.0832, "high": 1.0851, "low": 1.0828, "close": 1.0844, "volume": 1247 },
      ...
    ],
    "indicators": {
      "rsi":         true,
      "rsi21":       { "type": "rsi",   "length": 21 },
      "stochFast":   { "type": "stoch", "k": 5,  "d": 3, "smoothK": 3 },
      "stochSlow":   { "type": "stoch", "k": 21, "d": 7, "smoothK": 5 },
      "macd":        true,
      "orderBlocks": true,
      "fvg":         true
    }
  }'

The indicators object — the whole point

Each entry maps an output name (the key) to a spec:

• true — run the indicator with default params; key doubles as the type.
• { ...params } — params object; missing type falls back to the key.
• { "type": "<name>", ...params } — run a known indicator under a custom output name. This is how you stack multiple instances of the same indicator (e.g. four stochs with different params, three EMAs at different lengths).

"indicators": {
  "rsi":    true,
  "rsi21":  { "type": "rsi",   "length": 21 },
  "stochA": { "type": "stoch", "k": 5,  "d": 3 },
  "stochB": { "type": "stoch", "k": 21, "d": 7 }
}

The response contains rsi, rsi21, stochA, stochB. Nothing else. Duplicate output names are physically impossible by JSON-object construction — you can't shoot yourself in the foot with this API even if you try.

Concepts

The outputs below map to a handful of recurring trading ideas. If you've used any TradingView-style charting tool most of these will be familiar; if not, the short framing here is enough to pick the right output for the job.

Series vs events. A Series output is one value per input bar (warmup positions are null) — these are continuous quantities you can chart. An event output is a sparse array of objects pinpointing things that just happened (a swing, a block, a structure break). Series tell you state; events tell you occurrences.

Primitives only — no signals. Wickworks does not emit interpretive signals (no divergence detection, no MA-cross events, no "buy/sell" tags). Everything returned is either a raw indicator series, a structural fact (an order block was formed at this bar, price closed past this swing), or a pre-baked summary over those — never a judgment about what to do. If you want divergences, MACD-cross events, golden/death crosses, or any other derived signal, build that layer in your own consumer.

The four questions every indicator answers part of:

1. What's the trend? → Moving averages, ADX, supertrend, ichimoku.
2. Is momentum behind it? → RSI, MACD, stochastic, MFI.
3. How much room is there? → ATR, Bollinger Bands, Donchian, Keltner.
4. Is volume backing it up? → OBV, CMF, A/D, KVO, VWAP.

In-house event constructs you won't find on TradingView:

• srLevels uses at most the latest 500 bars, a fixed five-bar separation between distinct touches, and at least two tests within half an ATR. It returns up to three nearest levels on each side of price.
• Order blocks and FVGs expose different state. Order blocks include mitigation state and times so consumers can choose their own freshness rule. FVG output contains only gaps that remain unmitigated.
• BOS vs CHoCH are classifications produced by the configured Smart Money Concepts swing algorithm. A BOS marks a same-direction structural break. A CHoCH marks an opposite-direction break. Neither label proves continuation or reversal.

Available indicators

Want the formal contract for tooling / validators? See schema.json — full JSON Schema Draft 2020-12. The reference below is the human version: categories + per-indicator blurbs + params tables + return shapes + examples.

Three series shapes are shared across most outputs:

• Series — one value per bar (number | null), aligned 1:1 with input bars. Warmup positions are null.
• FlagSeries — one 0/1 integer per bar.
• DirectionSeries — one -1/+1 integer per bar (1 = bullish/long, -1 = bearish/short).

Indicators below are grouped by what they tell a trader, not by parameter shape. Each subsection starts with a one-paragraph framing of the category, then lists every indicator in it with a short "what it is / when to use it" blurb.

Moving averages — trend bias and dynamic levels

Smoothed price lines. Each flavor trades responsiveness against lag differently — pick by how fast you want the curve to react to new bars. Trader use: define the dominant direction (price above/below the MA = bull/bear bias), identify dynamic support/resistance the market keeps touching, fire crossover signals (fast MA crossing slow MA = trend shift).

All take a single length parameter and return a Series.