data-pipeline

name	data-pipeline
description	Run an end-to-end data analysis pipeline — connect, profile, analyze, and report. Use when performing a complete analysis workflow.
allowed-tools	mcp__localdata__connect_database mcp__localdata__describe_database mcp__localdata__execute_query mcp__localdata__get_data_quality_report mcp__localdata__analyze_hypothesis_test mcp__localdata__analyze_regression mcp__localdata__analyze_clusters mcp__localdata__reduce_dimensions mcp__localdata__analyze_time_series mcp__localdata__forecast_time_series mcp__localdata__analyze_ab_test mcp__localdata__analyze_effect_sizes mcp__localdata__export_schema mcp__localdata__export_structured
argument-hint	<source-path> <analysis-type: regression|clustering|forecast|correlation|ab-test>

Data Pipeline

Run a complete analysis workflow from connection through profiling, analysis, and structured reporting.

Steps

1. Parse arguments. Extract the source path and analysis type from $ARGUMENTS. The first argument is the file path or connection string. The second argument is the analysis type: one of regression, clustering, forecast, correlation, or ab-test.

2. Connect and profile. Call connect_database with the source path. Then call describe_database to get the schema and get_data_quality_report to assess data quality. Summarize table structure, row counts, and quality scores.

3. Assess data readiness. Review the quality report. If critical issues exist (more than 30% nulls in key columns, severe duplicates), note them as caveats. Identify the columns relevant to the requested analysis type.

4. Route to the appropriate analysis. Based on the analysis type argument:

   regression -- Call analyze_regression with the target column and feature columns. Then call execute_query if needed to inspect residuals. Report model coefficients, R-squared, and feature importance.

   clustering -- Call analyze_clusters with numeric feature columns and algorithm "kmeans". Try k=2 through k=5 and compare silhouette scores. Call reduce_dimensions with PCA for a 2D summary. Report cluster profiles and quality.

   forecast -- Call analyze_time_series to decompose the series and test stationarity. Then call forecast_time_series with the target column and desired horizon. Report trend, seasonality, and forecast values with confidence intervals.

   correlation -- Call analyze_hypothesis_test with test type "correlation" for each numeric column pair. Call analyze_effect_sizes for the strongest relationships. Report a ranked correlation matrix with significance levels.

   ab-test -- Call analyze_ab_test with the metric and group columns. Call analyze_effect_sizes to quantify practical significance. Report group comparison, p-value, effect size, and ship/iterate/no-ship recommendation.

5. Export results. Call export_schema to capture the data structure. If the analysis produced structured output suitable for downstream use, call export_structured to save it in a portable format.

6. Present the report. Deliver a structured summary covering:

   • Data source: file type, tables, row and column counts
   • Quality assessment: completeness, consistency, issues found
   • Analysis results: key findings from the selected analysis type
   • Confidence: statistical significance levels and effect sizes where applicable
   • Limitations: data quality caveats, sample size concerns, assumption violations

7. Recommend follow-up. Suggest complementary analyses:

   • After regression: /localdata-mcp:analyze-correlations to validate predictor selection
   • After clustering: /localdata-mcp:regression using cluster labels as features
   • After forecast: /localdata-mcp:analyze-correlations to find external predictors
   • After correlation: /localdata-mcp:regression with the strongest predictors
   • After ab-test: /localdata-mcp:cluster-analysis to find segment-level effects