pdf-parse-mcp
@ai-zen/pdf-parse-mcp is an MCP server (stdio) that takes PDFs apart precisely for agents: it returns the coordinates and sizes of text, vector shapes and bitmaps, plus on-demand raster rendering.
It is not about "flattening a PDF into a blob of plain text". It is built for the "reconstruct a design file from a PDF" use case: you get the position, size, color and stroke width of every glyph, line and rectangle in the file.
- Precise geometry: text carries a baseline origin and bbox, font size, rotation and color; shapes carry fill/stroke, alpha, line width and boxes; bitmaps carry their placement box and intrinsic pixel size.
- Screen coordinate system: everything uses a top-left origin, y down, in points (1pt = 1/72 inch) — the intuitive coordinate system of screens and design tools, instead of PDF's internal "bottom-left origin, y up".
- On-demand rendering: rasterize a whole page or any region to PNG/JPEG;
regionplus a largescaleis a magnifying glass. - Single source of truth: this repo's
docs/is the only source of documentation; the website only aggregates and renders it.
The three tools
| Tool | Purpose |
|---|---|
pdf_info | Document-level info: page count, per-page size and rotation, metadata |
pdf_extract | The core tool: extract text / vector shapes / bitmaps from one page |
pdf_render | Rasterize one page to an image (whole page or a region) |
Documentation navigation
- Getting Started — Requirements, install, wiring an MCP client, and your first call.
- MCP Tools — Full parameters and return shapes for
pdf_info/pdf_extract/pdf_render. - Core Concepts — Coordinate convention, hidden text layers, patterns and gradients, and gotchas.
- API Reference — Programmatic API
openPage/extractPage/renderPage/renderOverlay. - Reconstructing a Design from a PDF — The recommended workflow and tips.
- Overlay Debug Images — Visually verify extraction accuracy with an overlay.