Read relevant PDF passages before filling agent context

Search and read selected PDF pages to reduce the document text sent to a language model.

1. Inspect page count and warnings with inspect_pdf.
2. Use get_document_map and search_document to locate candidate evidence.
3. Read only the selected pages; verify important quotes with verify_document_citation.
4. Send selected text and page evidence to your model, widening retrieval if needed.

The native-text tools do not invoke a language model. Smaller retrieved inputs can reduce model context consumption, but savings depend on the document and workflow. A search hit is not proof that every relevant passage was found; coverage matters more than minimizing tokens.

Start with the included tools

Indexing a PDF uses the published x402 document price. Retrieval and verification are included with the returned short-lived bearer capability; no paid generation is required for these workflows.

Copy the REST example · Connect an MCP client · Review limits and retention