PDFs are converted to per-page JPEG images instead of using document content blocks

Problem

When a PDF is sent to Claude Code, each page is converted to a separate JPEG image and sent as individual image content blocks with type: "base64". This means a 20-page PDF counts as 20 images against the API's 100-image limit.

The Anthropic API supports a document content block type where a PDF is sent as a single document. The Claude desktop/web app uses this, so a PDF counts as 1 document rather than N images.

Impact

• A conversation with a few PDFs easily hits the too many images and documents: 105 + 0 > 100 error and becomes completely unusable
• In my case, ~5 PDF page batches (totaling 105 page-images) bricked the conversation
• The per-page JPEG conversion also bloats the conversation history file (my .jsonl went to 12MB, mostly base64 image data)
• Image quality is likely worse than native PDF rendering

Expected behavior

PDFs should be sent using the API's document content block type (like the desktop app does), so each PDF counts as 1 document rather than N images.

Reproduction

1. Send a 20+ page PDF to Claude Code (e.g. cat paper.pdf | claude or reference it in conversation)
2. Send a few more PDFs or images in the same conversation
3. Conversation fails with too many images and documents error

Workaround

Manually edit the conversation .jsonl file to remove image blocks, then resume the conversation.