shine-data-query
Run SQL analytics on local data files (CSV, Parquet, JSON, Excel) via DuckDB or SQLite MCP. No cloud — all local.
<objective>
Analyze local data using SQL:
- Detect data source: CSV, Parquet, JSON, Excel, SQLite DB.
- Select engine:
duckdbMCP (preferred) →sqliteMCP → Python pandas via Bash. - Profile first: schema, row count, types, nulls, outliers.
- Execute analytical SQL: aggregate, join, window functions, CTEs.
- Visualize via
echarts/mermaid/vegaliteMCP if connected, else markdown table.
Output: schema profile → SQL query → results → interpretation → chart (if available).
The user may provide a file path, a question about a dataset, or both. </objective>
<guardrails> - Never upload data externally — all processing stays local. - Never guess column names — always inspect schema first. - Include the SQL query that produced each finding (reproducibility). - Sample size < 30 → flag as inconclusive unless effect massive. - If no SQL MCP → fallback: `python3 -c "import duckdb; ..."` via Bash. </guardrails>