pdf-mcp

#!/usr/bin/env ruby

# Provides MCP for exploring pdf files.
# Make sure you install mupdf and pdfgrep (e.g. `brew install mupdf pdfgrep`).

require 'mupdf'
require 'pdf-reader'
require 'tiny_mcp'
require 'base64'

# Requires mupdf
class PdfPageCount < TinyMCP::Tool
  name 'PdfPageCount'
  desc 'Get page count of pdf files'
  arg :paths, :string,
    'Paths to pdfs separated by semicolons (e.g. "foo/bar.pdf;foo/baz.pdf")'

  def call(paths:)
    paths.split(';').map { |path|
      path.strip!
      reader = MuPDF::Document.new(path)
      "#{path}: #{reader.info.pages} pages"
    }.join("\n")
  end
end

# Requires pdfgrep
class PdfSearch < TinyMCP::Tool
  name 'PdfSearch'
  desc 'Search PDF files for text using regular expressions. Returns pages ' \
    'where the text was found.'

  arg :path, :string, 'Path to the PDF file'
  arg :regex, :string, 'Regular expression to search (e.g. "Chapter [0-9]+")'

  def call(path:, regex:)
    command = "pdfgrep --color never -inP #{regex.shellescape} #{path}"
    output = `#{command}`
    pages = output.split("\n").map { _1.split(':').first.to_i }.uniq
    pages.any? ? "Found on pages: #{pages.join(',')}" : 'Nothing was found'
  end
end

# Requires pdf-reader, mupdf
class PdfRead < TinyMCP::Tool
  name 'PdfRead'
  desc 'Read PDF files. Use format="image" if the page might have images, ' \
    'tables or diagrams.'

  arg :filepath, :string, 'Path to the PDF file'
  arg :pages, :string, 'List of pages to read (1-based, e.g. "1,2,5")'
  opt :format, :string, 'Get pages as text or image (e.g. "text" vs "image")'

  def call(filepath:, pages:, format: 'text')
    format = format == 'image' ? 'image' : 'text'
    pages = pages.split(',').map(&:to_i)

    format == 'image' ?
      read_images(filepath, pages) :
      read_text(filepath, pages)
  end

  private

  def read_text(path, pages)
    reader = PDF::Reader.new(path)
    pages.map { { type: 'text', text: reader.page(_1).text } }
  end

  def read_images(path, pages)
    reader  = MuPDF::Document.new(path)
    Dir.mktmpdir do |dir|
      pages.map {
        img = "#{dir}/page_#{_1}.png"
        reader.draw(page: _1, format: 'png', path: img)
        data = Base64.strict_encode64(File.binread(img))
        { type: 'image', mimeType: 'image/png', data: data }
      }
    end
  end
end

TinyMCP.serve(PdfPageCount, PdfSearch, PdfRead, server_name: 'pdf-mcp')