Transformers

Transformers

Transformers are a way to sanitize, transform and alter the data:

• going INTO a function (MCP, or Collection). Trigger: onFunctionInput.
• coming OUT a function (MCP, or Collection). Trigger: onFunctionOutput.
• being loaded from a resource. Trigger: onResource.

Transformers are defined outside the realm of sessions, and are triggered by hooks.

Operators

• parser: (csv/json) if the input data is plain text parse the input data into a structured output.
• jsonata: apply a JSONata expression to the structured input (currently deprecated)
• jmesPath: apply a JMESPath expression to the structured input
• expr: apply an Go Expr expression to the structured input
• code: run a script against the input data, assuming the integrating application provides a scripting engine. The CLI supports a vanilla JavaScript runtime.

A single transformer can implement all the above operators and will be applied as a transformation chain top-down.

YAML

This example sanitizes GitHub pull request data (MCP function):

transformers:
  - name: cleanup_atlassian
    onFunctionOutput: search_pull_requests
    jsonata: |-
      items.{
        "title": title,
        "state": state,
        "url": html_url,
        "author": user.login
      }

This example loads a csv resource into memory, but first makes into a data structure so that the prompt can reference it:

transformers:
  - name: csv1
    onResource: data/data2.csv
    parser: csv
    jmesPath: '[1:].{first_name: [0], last_name: [1]}'
sessions:
  parsed_content_into_memory:
    prompt: |-
      who was {{ (index .vars.data2 2).first_name }} {{ (index .vars.data2 2).last_name }}.
    resources:
      - identifier: data/data2.csv
        in: vars
        var: data2

Go

Same as the YAML version, nothing special here.