classification
Configure classification parameters, input/output bindings and lineage.
Classify rows into a bounded set of business labels using model reasoning.
Install
Install the Transforms bundle, then select the classification capability on a pipeline transform node. Runtime entrypoint: transforms/classification.ts. The installed capability must be enabled and pass preflight.
Configure
Execution: AI-based. Declared runtime support: supported.
Parameters
| Field | Type | Required | Details |
|---|---|---|---|
sourceFields | array | Yes | Fields to combine and classify. Enter one field per line. Items: string. |
labels | array | Yes | Available labels for the classifier. Items: object. |
instructions | string | No | Optional domain-specific guidance for the classifier. |
multiLabel | boolean | No | When true, output a JSON array of labels instead of a single label. Default: false. |
outputField | string | Yes | Primary output field that will store the result. |
confidenceField | string | No | Field that stores model confidence for each row. |
rationaleField | string | No | Field that stores brief reasoning for the row-level decision. |
View the complete contract for nested parameter shapes, defaults, named bindings and lineage.
Bindings
- Input
in: Input (required) - Output
out: Output
Verify the result
Create a small input with known values that exercise classification and supply the required parameters above. Validate, save and run the pipeline. Inspect the output records and compare row counts and changed fields with the input. Check the run’s errors and evidence before accepting a conversational summary.
Review the confidence and rationale fields where supplied. Include ambiguous examples and inspect the result; model output is not an observed business fact.
Lineage and changes
Lineage mode: row_preserving. This capability declares one output row per input row in the same order. Verify that downstream evidence still refers to the correct source row.
Changing parameters changes the computation. Review the saved pipeline revision before rerunning it. Publishing a new plugin version and editing node parameters are different changes.
FAQ
Why does validation fail?
Check required parameters, the installed capability version, runtime support and the named input/output bindings. A missing nested field may be visible only in the full contract.
Is this a standalone MCP tool?
It is a plugin capability used by a pipeline, not a new top-level MCP tool with this operation name. An assistant can inspect, validate and execute the pipeline through the MCP reference.