Ranked drivers are common; the second layer is what earns its place. In the project's Titanic example sex is the top driver — women survived far more often than men — and within women, third class reverses the picture. A flat importance list hides that reversal.
A key-driver and feature-importance server built on araxai's CleverMiner association rule analysis. Load a CSV into session memory, name the outcome, and get factors ranked against the baseline rate, with sub-driver analysis showing how factors combine, amplify or reverse each other.
- A CSV loaded into session memory, and the loaded datasets listed
- Key drivers of a target outcome, each with its rate against the baseline
- Sub-drivers returned automatically, showing how factors behave within a group
- Driver analysis conditioned on a segment variable, against that segment's own baseline
- Filters that narrow the analysis to a subgroup and re-rank inside it
Nothing to install — uvx fetches and runs the package. Without uv, pip install uv or pipx install key-drivers-mcp. It reads CSV files you already have; no account, no key.
One command — pipx install key-drivers-mcp