pdf-mcp

PDF tools for Claude — an MCP server, CLI, and Claude Code skill for reading and writing PDF files.

Stack: Mozilla PDF.js (pdfjs-dist) for text extraction · pdf-lib for creating and manipulating PDFs · @modelcontextprotocol/sdk for the MCP server.

No build step — plain JavaScript ESM, runs directly with Node.js 18+.

MCP Server

Setup

Clone the repo and link it globally so pdf-mcp is on your PATH:

git clone https://github.com/angshuman/pdf-mcp.git
cd pdf-mcp
npm install
npm link          # registers the pdf-mcp command globally

Claude Code:

claude mcp add pdf-mcp -- pdf-mcp

Claude Desktop — add to %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS):

{
  "mcpServers": {
    "pdf-mcp": {
      "command": "pdf-mcp"
    }
  }
}

If you'd rather not use npm link, point directly at the script:

claude mcp add pdf-mcp -- node /path/to/pdf-mcp/src/server.js

{
  "mcpServers": {
    "pdf-mcp": {
      "command": "node",
      "args": ["/path/to/pdf-mcp/src/server.js"]
    }
  }
}

Tools

CLI

Install

npm install
npm link          # makes pdf-tool available globally

Usage

# Extract text
pdf-tool read report.pdf
pdf-tool read report.pdf --page 3
pdf-tool read report.pdf --out extracted.txt

# Show metadata
pdf-tool info report.pdf

# Create a PDF from a text file or stdin
pdf-tool write out.pdf --in content.txt --title "My Doc" --author "Jane"
cat content.txt | pdf-tool write out.pdf

# Merge PDFs
pdf-tool merge combined.pdf a.pdf b.pdf c.pdf

# Extract specific pages (1-based, comma-separated)
pdf-tool extract input.pdf output.pdf 1,3,5

# Split into individual page files
pdf-tool split input.pdf ./pages/

# Render a single page to an image (PNG or JPEG)
pdf-tool image input.pdf page1.png --page 1
pdf-tool image input.pdf page1.jpg --page 1 --scale 3

# Render all pages to images in a directory
pdf-tool images input.pdf ./images/
pdf-tool images input.pdf ./images/ --format jpeg --scale 1.5

Scale controls DPI: 1.0 = 72 DPI · 2.0 = 144 DPI (default) · 3.0 = 216 DPI

Claude Code Skill

The /pdf slash command is in .claude/commands/pdf.md and is available automatically within this project.

/pdf read ./report.pdf
/pdf write ./out.pdf summarize the meeting notes
/pdf merge ./combined.pdf a.pdf b.pdf
/pdf extract ./in.pdf ./out.pdf 1,3,5
/pdf split ./in.pdf ./pages/

Tests

Uses the Node.js built-in test runner — no extra dependencies.

npm test

40 tests across 8 suites covering all operations: write, read, info, merge, extract pages, split, page-to-image, and pdf-to-images.

Notes

• pdfjs-dist emits stderr warnings about LiberationSans.ttf and glyph paths when rendering PDFs that use non-embedded standard Type1 fonts (Helvetica, Times, etc.). These are cosmetic — text extraction is unaffected, and image rendering works correctly for PDFs with embedded fonts (the common case for PDFs from Word, Adobe, Google Docs, etc.).

• All file paths passed to the MCP tools and CLI can be absolute or relative to the current working directory.

• Page numbers are always 1-based in both the MCP tools and CLI.