pdf-mcp-server

An MCP server for intelligent PDF reading and retrieval from an Obsidian vault. Solves the problem of large PDFs exceeding context window limits by providing chunked, searchable access.

Features

• List PDFs — Discover all PDFs in your vault
• PDF Info — Get metadata (pages, author, title)
• Table of Contents — Extract TOC/bookmarks for structure-aware reading
• Read by Pages — Read specific page ranges (1–20 pages per call)
• Search — Find keywords with surrounding context snippets
• Read by Chapter — Navigate TOC chapters automatically

Setup

pip install -r requirements.txt

Configuration

Set the vault path via environment variable:

export PDF_MCP_VAULT_PATH="D:\claude code\Angel"

Or it defaults to D:\claude code\Angel.

MCP Config (Claude Code)

{
  "mcpServers": {
    "pdf-reader": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "src.server"],
      "cwd": "D:\\claude code\\pdf-mcp-server",
      "env": {
        "VAULT_PATH": "D:\\claude code\\Angel"
      }
    }
  }
}

Usage

1. Call list_pdfs to see available PDFs
2. Call get_pdf_toc to understand document structure
3. Use read_pdf_chapter for section-based reading
4. Use search_pdf to find specific topics
5. Use read_pdf_pages for precise page ranges