Claude Analyze Document From URL

arazzo: 1.0.1
info:
  title: Claude Analyze Document From URL
  summary: Count tokens for a URL-referenced document, then ask Claude a question about it.
  description: >-
    A document-grounded question-answering flow. This spec does not expose a
    Files API, so the common "upload a file then reference it in a message"
    pattern is adapted to use a URL-referenced document content block, which the
    Messages API supports natively. The workflow first counts the tokens the
    document plus question would consume, then sends a message containing a
    document block (pointing at the document URL) and a text question so Claude
    can answer grounded in the document. Every step spells out its request
    inline — including the required x-api-key and anthropic-version headers — so
    the flow can be read and executed without opening the underlying OpenAPI
    description.
  version: 1.0.0
sourceDescriptions:
- name: claudeApi
  url: ../openapi/claude-messages-api.yml
  type: openapi
workflows:
- workflowId: analyze-document-from-url
  summary: Token-check then answer a question about a URL-referenced document.
  description: >-
    Counts tokens for a document-plus-question payload, then sends a message
    with a URL document block and a text question, returning the grounded answer.
  inputs:
    type: object
    required:
    - apiKey
    - model
    - documentUrl
    - question
    properties:
      apiKey:
        type: string
        description: Anthropic API key sent in the x-api-key header.
      anthropicVersion:
        type: string
        description: Value for the required anthropic-version header.
        default: '2023-06-01'
      model:
        type: string
        description: The model to analyze the document with.
      documentUrl:
        type: string
        description: The URL of the PDF or text document to analyze.
      question:
        type: string
        description: The question to ask about the document.
      maxTokens:
        type: integer
        description: Maximum number of tokens to generate in the answer.
        default: 1024
  steps:
  - stepId: countDocumentTokens
    description: >-
      Count the tokens the document and question will consume so the request can
      be verified to fit within the model's context window before answering.
    operationId: countMessageTokens
    parameters:
    - name: x-api-key
      in: header
      value: $inputs.apiKey
    - name: anthropic-version
      in: header
      value: $inputs.anthropicVersion
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.model
        messages:
        - role: user
          content:
          - type: document
            source:
              type: url
              url: $inputs.documentUrl
          - type: text
            text: $inputs.question
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      inputTokens: $response.body#/input_tokens
  - stepId: answerFromDocument
    description: >-
      Send the document block and question to the model and capture the grounded
      answer.
    operationId: createMessage
    parameters:
    - name: x-api-key
      in: header
      value: $inputs.apiKey
    - name: anthropic-version
      in: header
      value: $inputs.anthropicVersion
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.model
        max_tokens: $inputs.maxTokens
        messages:
        - role: user
          content:
          - type: document
            source:
              type: url
              url: $inputs.documentUrl
          - type: text
            text: $inputs.question
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      messageId: $response.body#/id
      content: $response.body#/content
      stopReason: $response.body#/stop_reason
      usage: $response.body#/usage
  outputs:
    estimatedInputTokens: $steps.countDocumentTokens.outputs.inputTokens
    messageId: $steps.answerFromDocument.outputs.messageId
    content: $steps.answerFromDocument.outputs.content
    usage: $steps.answerFromDocument.outputs.usage