PDF MIME Type

What does application/pdf mean?

The application/pdf media type identifies a document written in Portable Document Format. PDF was designed for dependable, device-independent presentation: a page can combine text, fonts, vector graphics, raster images, annotations, forms, links, metadata, and even embedded files while retaining its intended layout.¹

The registered extension is .pdf. A PDF starts with the ASCII characters %PDF- followed by a version number, such as %PDF-1.7 or %PDF-2.0. In hexadecimal, that common five-byte signature is 25 50 44 46 2D.¹

How PDF became an open standard

Adobe first published the PDF specification in 1993. Adobe maintained eight editions, from PDF 1.0 through PDF 1.7, before PDF 1.7 became ISO 32000-1 in 2008. PDF 2.0 is standardized as ISO 32000-2, whose current second edition was published in 2020.²³

The MIME registration has its own history. RFC 8118 updated the registration in 2017 and replaced RFC 3778. It defines application/pdf as a binary media type with no required or optional parameters, one registered extension, and no deprecated aliases.¹

PDF version numbers describe revisions of the base syntax. They are different from specialized conformance standards:

These formats are constrained uses of PDF, not alternate MIME types. A conforming PDF/A or PDF/UA file is still normally served as application/pdf.¹⁴

How to open a PDF

Current desktop browsers can display ordinary PDFs directly. Chrome and Firefox document built-in PDF viewers, and Chromium-based Edge and Opera provide the same basic path. Safari uses the PDF support built into Apple platforms.¹⁵⁶

For desktop use:

• Adobe Acrobat and Acrobat Reader open PDFs and provide navigation, search, forms, signatures, and annotation tools.⁷
• Apple Preview opens PDFs on macOS and supports viewing, search, markup, forms, and export.⁸
• A browser is usually sufficient for reading, printing, or downloading a straightforward document.

Text PDF versus scanned PDF

A page that looks like text may contain actual character data, a photograph of text, or both. That distinction matters:

• A born-digital PDF usually has a text layer that viewers can select, copy, search, and expose to assistive technology.
• A scanned PDF may contain only page images. It needs optical character recognition (OCR) before its words become searchable.
• A tagged PDF adds logical structure such as headings, reading order, lists, and alternate text. Visual fidelity alone does not guarantee accessibility.

PDF/A does not guarantee extractable text either. The Library of Congress notes that a conforming PDF/A file may consist of scanned page images and that structural tagging depends on the creation process.⁹

Support in Poly

Poly has a dedicated PDF pipeline rather than treating .pdf as generic binary data. The implementation recognizes both the application/pdf value and the .pdf extension, then:

• Renders pages and creates thumbnails for a page-by-page preview.
• Extracts PDF metadata including format, encryption, author, title, producer, creator, subject, and keywords when present.
• Extracts text from each page and invokes OCR when a page has no embedded text.
• Builds per-page visual embeddings, so diagrams and page appearance can participate in similarity search.
• Stores page-numbered full text so search results, poly file read, and agent tools can return the relevant pages.

Poly's viewer is built on PDF.js and can jump to a matched page. Agent reads interleave each selected page's extracted text with its rendered image. That preserves visual evidence such as layout, tables, figures, handwriting, and stamps that plain text extraction can miss.

Two processing limits are important. Poly does not index PDFs larger than 1 GiB, and page rendering and embeddings are capped at 1,024 pages. A single CLI or MCP read accepts at most 100 selected pages, so an agent must page through a longer document deliberately.

Converting a PDF

Choose the output by what you need to preserve:

• PDF to text or Markdown for quoting, search, and language-model workflows. Expect cleanup when reading order or tables are complex.
• PDF to PNG or JPEG for a fixed visual snapshot of each page. The result is no longer selectable or structurally accessible unless separate text is retained.
• PDF to Word or another editable document when revision matters more than exact pagination. Fonts, columns, footnotes, and forms may require manual repair.
• Office document to PDF for stable sharing or printing. Keep the editable source because PDF is optimized for final-form presentation rather than round-trip editing.
• PDF to PDF/A for an archival workflow. Use a validating converter; renaming a file or merely resaving it does not establish conformance.

Conversion cannot recover information that was never embedded. OCR can estimate text in a scan, but it cannot reconstruct the original fonts, document structure, formulas, or source application.

PDF security and privacy

PDF is an active container, not merely a stack of page images. It can include JavaScript, forms, links, multimedia, attachments, encryption, and actions that ask a viewer to open other material. RFC 8118 warns implementers to isolate scripts and executable content and to handle external links, embedded files, and decompression carefully.¹

For an untrusted PDF:

• Keep the browser, operating system, and PDF application updated.
• Avoid opening unexpected attachments or approving script-driven actions.
• Treat password protection as access control, not proof that a document is trustworthy.
• Inspect document metadata before publishing. Author, title, creator, keywords, timestamps, and other workflow details may travel with the file.
• Use a validator when archival, accessibility, or print-standard conformance matters.

Is PDF the same as application/octet-stream?

No. application/pdf states that the payload uses PDF syntax. application/octet-stream is a generic label for arbitrary binary data. A server that sends a valid PDF as application/octet-stream may cause a browser to download it instead of opening it inline, while changing the label does not turn non-PDF bytes into a PDF.

Likewise, renaming report.bin to report.pdf changes only the filename. A real PDF needs a valid structure, not just the extension or %PDF- prefix.