xai-toolkit
ML model explainability as plain-English narratives, exposed via MCP.
What It Does
Ask a question in VS Code Copilot:
"Why was sample 42 classified as malignant?"
Get back a deterministic English explanation:
"The model classified sample 42 as malignant (probability: 0.91) primarilybecause of three factors: worst_radius is 2.1× above average (pushing riskup by +0.28), worst_concave_points is elevated (+0.19), and mean_concavityexceeds the norm (+0.14)."
No plots to interpret. No code to run. English that a decision-maker can act on.
Quick Start
uv sync # Install dependencies
uv run pytest # Run tests
uv run python -m xai_toolkit.server # Start MCP server
Architecture
See AGENTS.md for full project structure and design decisions.See docs/decisions/ for Architecture Decision Records.
Design Principle
The LLM is the presenter, not the analyst. All computation and narrativegeneration is done deterministically in Python. The LLM chooses the right tooland presents the result — nothing more.
