Connect to a dataset, profile its structure and quality, and recommend what analyses to run next.
-
Connect to the data source. Call connect_database with the path or connection string from $ARGUMENTS. Note the assigned database name in the response.
-
Describe the schema. Call describe_database with the database name. Record the list of tables, column names, and column types. Identify which columns are numeric, categorical, datetime, and text.
-
Sample rows from each table. For each table (or the first 3 if many), call execute_query with SELECT * FROM <table> LIMIT 10. Inspect the returned rows to understand value ranges, formats, and potential join keys.
-
Describe key tables. For the most important tables (largest or most-referenced), call describe_table to get detailed column statistics including cardinality, null counts, and value distributions.
-
Run a quality report. Call get_data_quality_report with the database name. Review completeness, uniqueness, and consistency scores. Flag columns with high null rates (above 20%) or low cardinality that may need attention.
-
Summarize findings. Present a structured summary:
- Number of tables, total rows, and columns
- Data types breakdown (numeric, categorical, datetime, text)
- Quality issues found (nulls, duplicates, inconsistencies)
- Key relationships between tables (shared column names)
-
Recommend next analyses. Based on data characteristics, suggest specific next steps:
- Numeric pairs with potential relationships: suggest
/localdata-mcp:analyze-correlations
- Datetime column with a metric: suggest
/localdata-mcp:forecast
- Many numeric features: suggest
/localdata-mcp:cluster-analysis
- Target variable present: suggest
/localdata-mcp:regression
- Treatment/control groups: suggest
/localdata-mcp:ab-test
- Graph or network file: suggest
/localdata-mcp:graph-explore
