1
Open PDF
pypdfium2 renders pages while pdfplumber reads text, captions, tables, and coordinates for comparison.
PDF Card MCP
How a dense PDF becomes a readable card reader
The tool keeps the PDF local, renders visual material as images, extracts body text into cards, and then removes text that already belongs inside a table, figure, or formula crop.
2
Render Pages
Each source page becomes an embedded PNG for the source-page modal.
3
Find Assets
Captions, table boxes, embedded images, vector regions, and display equations become image crops.
4
Suppress Overlap
Text blocks inside asset regions are removed from body cards so diagram labels do not duplicate.
5
Merge Text
Nearby body blocks are joined before splitting into digestible cards.
6
Write HTML
The reader is one standalone file with embedded images, search, navigation, and font-size controls.
Where quality goes wrong
PDFs do not expose clean article structure. A visual chart may be stored as vector lines plus many tiny text blocks, while a table may not expose a reliable table box.
Fallback crop can grab nearby prose
What the reader tries to do
Keep tables, figures, and formulas visible as images, but use geometry to avoid duplicating their internal labels as separate cards.
FigureFigure card
One crop with caption and source-page link.
FormulaFormula card
One tight equation crop so math layout stays faithful.
Body card
Only prose outside the asset crop region.
Best case
The PDF has a text layer, good captions, and detectable table boxes. Crops are tight and cards read cleanly.
Hard case
The PDF has vector diagrams, multi-column flow, or table text without borders. The converter has to infer boundaries.
User control
The generated reader includes a font-size slider, source-page modal, and standalone embedded assets.