pdf-toolkit-mcp

A comprehensive MCP (Model Context Protocol) server for PDF operations. 37 tools for reading, searching, creating, merging, splitting, watermarking, stamping, form filling, redacting, annotating, comparing, optimizing, and more.

Built on open-source libraries — not affiliated with Adobe. Uses pdf-lib for PDF manipulation and unpdf (Mozilla pdf.js) for text extraction.

Requirements

• Node.js >= 18 (tested on Node 18, 20, 22, 24)

• npm or compatible package manager

Installation

git clone https://github.com/beepboop2025/pdf-toolkit-mcp.git
cd pdf-toolkit-mcp
npm install
npm run build

Configuration

Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "pdf-toolkit": {
      "command": "node",
      "args": ["/path/to/pdf-toolkit-mcp/dist/index.js"]
    }
  }
}

Dev mode (no build step)

{
  "mcpServers": {
    "pdf-toolkit": {
      "command": "npx",
      "args": ["tsx", "/path/to/pdf-toolkit-mcp/src/index.ts"]
    }
  }
}

Tools Overview

Tool Reference

Reading

pdf_read_text — Extract text from all or specific pages.

• filePath (string, required) — PDF path

• pages (number[], optional) — Page numbers to extract (1-indexed). Returns per-page text with headers.

pdf_info — Page count, file size, all metadata, form field count, page dimensions (pts + inches), rotation, PDF version.

• filePath (string, required) — PDF path

• password (string, optional) — For encrypted PDFs

Searching

pdf_search — Full-text search across all pages. Returns matches with page numbers and surrounding context.

• filePath (string, required) — PDF path

• query (string, required) — Text or regex pattern

• caseSensitive (boolean, default: false)

• regex (boolean, default: false) — Treat query as regex

• contextChars (number, default: 60) — Characters of context around each match

pdf_compare — Compare text content of two PDFs page by page. Reports identical/different pages with word-level change summary.

• filePath1, filePath2 (strings, required)

Creating

pdf_create — Create a PDF from text with auto word-wrapping, pagination, and headings.

• outputPath, content (required) — Lines starting with # render as bold headings

• fontSize (default: 12), margin (50), lineSpacing (1.4)

• pageSize: A4, Letter, Legal, A3, A5

• font: Helvetica, TimesRoman, Courier

• title, author (optional metadata)

pdf_images_to_pdf — Convert PNG/JPG images to a PDF, one per page.

• imagePaths (string[], required), outputPath (required)

• pageSize: A4, Letter, Legal, FitImage (default: A4)

• margin (default: 0) — Padding around image

Manipulating

pdf_merge — Merge multiple PDFs.

• filePaths (string[], min 2), outputPath

pdf_split — Split into individual pages or custom ranges.

• filePath, outputDir

• ranges (optional) — [{start, end}] (1-indexed, inclusive). Omit to split into individual pages.

pdf_rotate — Rotate pages 90/180/270 degrees clockwise.

• filePath, outputPath, rotation ("90"/"180"/"270")

• pages (optional) — Omit for all pages

pdf_delete_pages — Remove specific pages.

• filePath, outputPath, pages (number[], required)

pdf_extract_pages — Extract specific pages into a new PDF.

• filePath, outputPath, pages (number[], required)

pdf_reorder — Reorder or duplicate pages.

• filePath, outputPath, pageOrder (number[]) — e.g. [3, 1, 2] or [1, 1, 2] to duplicate page 1

pdf_insert_pages — Insert pages from one PDF into another at a specific position.

• targetPath, sourcePath, outputPath

• insertAfter (number) — 0 = beginning, use target page count for end

• sourcePages (number[], optional) — Which source pages to take. Omit for all.

pdf_reverse — Reverse page order.

• filePath, outputPath

pdf_blank_pages — Insert blank pages after specified positions.

• filePath, outputPath

• afterPages (number[]) — e.g. [0, 3] inserts blanks at the beginning and after page 3

• pageSize: MatchPrevious (default), A4, Letter, Legal

pdf_crop — Crop pages by setting a crop box (does not resize, hides content outside the box).

• filePath, outputPath, x, y, width, height (all in points)

• pages (optional)

pdf_overlay_pdf — Overlay a single page from one PDF on top of another. Uses a single overlay page applied to selected base pages — not a page-by-page multi-page overlay.

• basePath (background), overlayPath (foreground), outputPath

• overlayPage (default: 1) — Which overlay page to use

• pages (optional) — Which base pages to apply it to

pdf_page_layout — Arrange multiple pages onto single sheets (N-up printing). Scales and positions source pages into a 2-up or 4-up grid.

• filePath, outputPath

• layout: "2-up" (2 pages side-by-side) or "4-up" (2×2 grid)

• pageSize: A4, Letter, Legal (default: Letter)

Overlays

pdf_watermark — Diagonal text watermark, properly centered regardless of rotation angle.

• filePath, outputPath, text (required)

• opacity (0.15), fontSize (60), color (#888888), rotation (45 degrees)

• pages (optional)

pdf_page_numbers — Add page numbers to every page.

• filePath, outputPath

• position: bottom-center (default), bottom-left, bottom-right, top-center, top-left, top-right

• startNumber (1), prefix (""), suffix (""), fontSize (10), margin (30), color (#4d4d4d)

pdf_add_text — Add text at exact coordinates. Supports multi-line (\n) and bold fonts.

• filePath, outputPath, text, page, x, y

• fontSize (12), color (black), lineSpacing (1.4)

• font: Helvetica, HelveticaBold, TimesRoman, TimesRomanBold, Courier, CourierBold

pdf_add_image — Add PNG/JPG at exact coordinates.

• filePath, outputPath, imagePath, page, x, y

• width, height (proportional scaling if only one given; original size if both omitted)

• opacity (1)

pdf_header_footer — Headers and/or footers with dynamic placeholders.

• filePath, outputPath

• header, footer — Use {page} for current page, {pages} for total

• fontSize (9), margin (30), align (center/left/right), color (#4d4d4d)

pdf_stamp — Styled business stamps with border and rotation.

• filePath, outputPath

• type: APPROVED, REJECTED, CONFIDENTIAL, DRAFT, FINAL, COPY, VOID, ORIGINAL, REVISED, FOR REVIEW, CUSTOM

• customText (string, only for CUSTOM type)

• position: top-right (default), top-left, bottom-right, bottom-left, center

• pages (optional), opacity (0.75), rotation (-15)

pdf_sign — Add a signature image with smart positioning.

• filePath, outputPath, imagePath

• page (default: last page)

• position: bottom-right (default), bottom-left, bottom-center

• width (150, height scales proportionally), margin (50)

Annotating

pdf_redact — Draw opaque rectangles over regions.

• filePath, outputPath

• regions — [{page, x, y, width, height}]

• color (default: black)

WARNING: Visual-only redaction. This tool covers content with an opaque box but does NOT remove the underlying text or data from the PDF structure. The original content can still be extracted programmatically by anyone with a PDF parser. Do NOT use this for true sanitization of sensitive data (SSNs, passwords, PII, etc). For secure redaction that strips content from the PDF, use Adobe Acrobat Pro, qpdf, or mutool clean.

pdf_highlight — Translucent highlight boxes (like a highlighter pen).

• filePath, outputPath

• highlights — [{page, x, y, width, height, color?}] (per-highlight color, default: yellow)

• opacity (0.35)

pdf_draw — Draw shapes: rectangles, lines, and ellipses.

• filePath, outputPath

• shapes — array of shape objects:

  • rectangle: {type: "rectangle", page, x, y, width, height, color?, borderColor?, borderWidth?, filled?, opacity?}

  • line: {type: "line", page, x, y, width, height} where x/y = start point, width/height = end point coordinates

  • ellipse: {type: "ellipse", page, x, y, width, height, color?, filled?, opacity?}

Forms

pdf_form_read — List all form fields with names, types, current values, and dropdown/radio options.

• filePath, password (optional)

pdf_form_fill — Fill form fields by name. Supports partial fills (unfilled fields remain editable).

• filePath, outputPath

• fields — {"fieldName": "value"}. Checkboxes: "true"/"false". Dropdowns/radio: the option value.

• flatten (default: false) — When true, all fields become non-editable (filled AND unfilled)

• password (optional)

Metadata

pdf_set_metadata — Update document metadata. Only specified fields change; others are untouched.

• filePath, outputPath

• title, author, subject, creator, producer (all optional strings)

• keywords (string[], optional)

Security

pdf_decrypt — Remove password protection (requires the correct password).

• filePath, password, outputPath

Optimize

pdf_flatten — Flatten all form fields into static page content. Fields become non-editable. Useful for finalizing documents.

• filePath, outputPath

pdf_compress — Reduce PDF file size by rebuilding document structure. Copies pages to a fresh PDF (dropping orphaned objects), optionally strips metadata and flattens forms. Results vary.

• filePath, outputPath

• stripMetadata (boolean, default: false) — Remove title, author, subject, keywords

• flattenForms (boolean, default: false) — Flatten form fields

pdf_validate — Health check for a PDF file. Checks if it can be parsed, reports page count, metadata, form fields, page dimensions, and any structural issues.

• filePath (string, required)

pdf_repair — Attempt to repair a damaged PDF by loading with lenient parsing, copying all recoverable pages to a clean document, and re-saving.

• filePath (damaged PDF path), outputPath

Attachments

pdf_attach — Embed files as attachments inside a PDF. Attached files travel with the PDF and can be extracted by PDF readers. Supports any file type with auto-detected MIME types.

• filePath, outputPath

• attachments — [{path, description?}] — Files to embed

• Recognized types: PDF, TXT, CSV, JSON, XML, HTML, PNG, JPG, GIF, SVG, ZIP, DOC/DOCX, XLS/XLSX (others default to application/octet-stream)

Error Handling

All tools return structured error messages. Common scenarios:

Tools never throw unhandled exceptions — all errors are caught and returned as MCP error responses with isError: true.

File Size & Performance

• No hard file size limit — bounded only by available memory (Node.js heap)

• PDFs up to ~100MB work reliably; larger files may need --max-old-space-size

• pdf_merge with many large files is the most memory-intensive operation

• pdf_read_text loads the entire PDF into memory for parsing

• All operations are single-threaded (no worker pool)

Named Colors Reference

All color parameters (color, borderColor, highlight color, etc.) accept:

Complete named color list (28 colors):

Coordinate System

• Origin: bottom-left corner of each page

• Units: points (1 inch = 72 points, 1 cm ~ 28.35 points)

• Standard page sizes:

Limitations

Tech Stack

• pdf-lib — PDF creation and manipulation (open-source, MIT)

• unpdf — Text extraction via Mozilla's pdf.js (open-source)

• @modelcontextprotocol/sdk — MCP server framework

Disclaimer

This project is not affiliated with, endorsed by, or associated with Adobe Inc. "PDF" is an open standard (ISO 32000). This tool uses open-source libraries to work with PDF files.

License

MIT