MCP Responses

Response handling

Read the same bounded JSON from content[0].text or structuredContent. Both channels contain agent-readable rows for MCP client compatibility.

Response modes

Mode	Behavior
`full`	The complete response fits within the configured MCP budget.
`agent_view`	Useful rows stay inline while bulky fields move to a stored result resource.
`preview`	A bounded preview is returned when the agent view still exceeds the budget.

Structured response

Example

{
  "success": true,
  "summary": "Scraped https://example.com",
  "data": { "markdown": "# Example Domain" },
  "next_actions": ["Use datablue_extract to structure the page"],
  "usage": { "credits_used": 1 },
  "response": {
    "mode": "agent_view",
    "truncated": true,
    "original_size_bytes": 120000,
    "returned_size_bytes": 18000,
    "max_bytes": 60000,
    "max_array_items": 10,
    "omitted_paths": ["$.data.raw_html"],
    "full_result": {
      "id": "018f7c2a-4b19-7d8e-a3f0-123456789abc",
      "uri": "datablue://results/018f7c2a-4b19-7d8e-a3f0-123456789abc/0",
      "mime_type": "application/json",
      "expires_at": "2026-08-26T10:30:00Z",
      "note": "Use datablue_get_result_items with this id for zero-credit logical paging."
    }
  }
}

Use response.truncated to detect omitted content. Call datablue_get_result_items with response.full_result.id, an offset, and a limit to read more stored rows at zero DataBlue credits. If exact raw reconstruction is required, start with response.full_result.uri, base64-decode each page, and follow next_page_uri until complete.