Ask a model for two hundred daily values with a weekly pattern and a slow upward trend and you get numbers that look right in a list and fall apart in a chart — the seasonality drifts, the trend is not monotonic, the noise is suspiciously uniform. Generating from an actual model of trend, seasonality and noise fixes the thing that was broken. The deliberate anomaly injection is what makes it useful for testing alerting and detection, where you need to know the ground truth. Presets are the practical touch: most people want realistic, not configurable, and having both means the quick path exists.
A synthetic time-series generator with configurable structure: trend, seasonality, noise, anomalies and multiple correlated streams, output in a form you can load.
- generate_timeseries produces a series with each component configurable rather than a plausible-looking list of numbers.
- Anomalies are injected deliberately, which is what makes the output usable for testing detection rather than only for filling a chart.
- Multiple correlated streams, so a dashboard with related metrics behaves the way a real one would.
- list_presets returns ready-made realistic configurations, so a usable series is one call rather than a parameter-tuning exercise.
- Output as JSON, CSV or SQL — the last one meaning it can go straight into a database.
Nothing to supply.
One command — npx -y timeweaver-mcp