Verify the OmniData database server is running, can reach DuckDB, and report its current guardrail configuration (row limits, timeout, DB path).

Use this first to confirm the MCP connection is alive.

List every table and view available in the database, with its type and column count. Call this first when exploring an unknown database -- run_sql_query needs real table names to work with.

Get the column names, types, and nullability for one dataset (table or view). Use list_datasets first to get valid names.

Args: dataset: Exact table or view name, as returned by list_datasets.

Get the exact row count for one dataset. Use this before run_sql_query on a large table to know how much data you're sampling from with a LIMIT.

Args: dataset: Exact table or view name, as returned by list_datasets.

Execute a read-only SQL query and return structured results.

Safety guardrails (see README "Design decisions"):

• Only SELECT / WITH / EXPLAIN / DESCRIBE / SHOW statements are permitted; DDL/DML and system commands are rejected.

• Only one statement per call.

• If the query has no LIMIT clause, one is auto-injected using row_limit (or the server default). Results are always hard-capped at the server's max_row_limit, regardless of what you request, to keep responses bounded.

• The query is cancelled if it runs longer than the server's configured timeout.

Use get_row_count first if you need to know the true size of a table beyond what this capped result shows.

Args: sql: A single read-only SQL statement. row_limit: Desired max rows (capped at the server's max_row_limit). Ignored if your query already has LIMIT.

Generate summary statistics for every column in a dataset: type, null percentage, approximate distinct count, min/max, and (for numeric columns) mean/stddev/quartiles. Uses DuckDB's built-in SUMMARIZE, so it runs efficiently even on large tables.

Args: dataset: Exact table or view name, as returned by list_datasets.

Run a read-only SQL query and render the result as a chart image (bar, line, or scatter). Subject to the same read-only safety guardrails as run_sql_query.

Returns an inline chart image on success. On failure (bad SQL, missing column, empty result, oversized render), returns an error dict instead -- check for an "error" key if the result isn't an image. (No static return-type annotation here: the mcp SDK's output-schema generation can't handle Image inside a Union type.)

Args: sql: A single read-only SQL statement producing the data to chart. Aggregate/group the data yourself for cleaner charts (e.g. GROUP BY category). chart_type: "bar", "line", or "scatter". x_column: Column name (from the query result) for the x-axis. y_column: Column name (from the query result) for the y-axis. series_column: Optional column to split into multiple series/ traces (e.g. one line per region). title: Optional chart title. Defaults to "{y_column} by {x_column}".

Run a declarative PySpark transformation pipeline against a DuckDB dataset -- for heavier aggregations/transformations than run_sql_query is meant for. NOT arbitrary code execution: each step must be one of a fixed set of operations, validated before running.

Supported operations (each a dict with an "op" key): {"op": "filter", "condition": ""} e.g. {"op": "filter", "condition": "revenue > 100"} {"op": "select", "columns": ["a", "b"]} {"op": "withColumn", "name": "new_col", "expression": ""} e.g. {"op": "withColumn", "name": "margin", "expression": "revenue - cost"} {"op": "groupBy_agg", "group_by": ["a"], "aggregations": {"b": "sum"}} aggregations map column -> function; functions: sum, avg, mean, count, min, max, stddev, variance {"op": "orderBy", "columns": ["a"], "ascending": true} {"op": "distinct"} {"op": "limit", "n": 100}

Steps run in the order given. A final row cap is always applied to the output regardless of what the pipeline itself requests.

Args: source_dataset: Exact table/view name, as returned by list_datasets. operations: Ordered list of pipeline steps (see above). row_limit: Desired max rows returned (capped at the server's max_row_limit).