PDF Memo MCP Server

A Model Context Protocol (MCP) server that lets Claude (or any MCP client) work with your PDFs safely. It can list PDFs, extract highlights with their exact text, return raw notes/annotations, read specific pages, and show metadata—all strictly limited to the directories you allow.

Features

• Annotation extraction: parse PDF annotations with sensible filtering (e.g., include Highlight, Text by default; ignore noisy Link/Popup unless requested).
• Highlight→text mapping: map highlight quads/rects to the underlying page text.
• Flexible page ranges: first, last, N, S-E, or omit for all.
• Metadata: title, author, subject, creator/producer, creation/mod dates.
• Configurable directories: allow multiple roots via positional args or repeated --allow-dir.
• Safety: access limited to allowed directories, file size/extension checks.

Requirements

• Python 3.10+ (3.11 recommended)
• pip install -r requirements.txt

PyPDF2>=3.0.0,<4
pdfplumber>=0.11.0,<0.12
mcp>=0.4.0

Installation

# 1) Clone this repo
# Windows PowerShell / macOS / Linux
cd <your-workspace>
git clone <this-repo-url> pdf-memo-mcp
cd pdf-memo-mcp

# 2) (Optional) Create a virtualenv
python -m venv .venv
# Windows
. .venv\Scripts\Activate.ps1
# macOS/Linux
source .venv/bin/activate

# 3) Install dependencies
pip install -r requirements.txt

Running the server

You can pass allowed directories either as positional args or with `--allow-dir` (repeatable). If you pass none, the server falls back to defaults: ~/Downloads, ~/Desktop, ~/Documents, and the current working directory.

# Example: two directories (mixed platforms shown for clarity)
python main.py ~/Downloads C:\\Temp

# Same via explicit flags (repeatable)
python main.py --allow-dir ~/Downloads --allow-dir C:\\Temp

# Optional tuning
python main.py ~/Downloads --max-file-size 52428800 --log-level DEBUG

Tip (Windows): If you run into import issues while developing, ensure your repo root is on PYTHONPATH, e.g. set PYTHONPATH=C:\\mcp-servers\\pdf-memo-mcp.

Claude Desktop integration

Create or update your claude_desktop_config.json to register this MCP server. Below is a minimal example.

{
  "mcpServers": {
    "pdf-memo": {
      "command": "C:/Users/yong/AppData/Local/Programs/Python/Python311/python.exe",
      "args": [
        "C:/mcp-servers/pdf-memo-mcp/main.py",
        "~/Downloads",
        "C:/Temp"
      ],
      "env": {
        "PYTHONPATH": "C:/mcp-servers/pdf-memo-mcp"
      }
    }
  }
}

macOS/Linux variant:

{
  "mcpServers": {
    "pdf-memo": {
      "command": "/usr/bin/python3",
      "args": [
        "/Users/you/mcp-servers/pdf-memo-mcp/main.py",
        "~/Downloads",
        "/Users/you/PDFs"
      ],
      "env": {
        "PYTHONPATH": "/Users/you/mcp-servers/pdf-memo-mcp"
      }
    }
  }
}

Notes

>

The server communicates over STDIO; Claude Desktop will handle transport automatically. You can add/remove directories later by editing args and restarting Claude Desktop.

Available tools (names & typical usage)

• extract_annotations(file_path, page_range=None, include_types="Highlight,Text", drop_empty=True)

Return raw annotations as JSON. Use include_types to include Link, Popup if you really need them.

• extract_highlights_with_context(file_path, page_range=None, drop_empty=True)

Return a JSON array of highlight contexts: {page, author, highlighted_text, note, position}.

• read_pdf_text(file_path, page_range=None)

Return JSON with extracted_pages (page text & char counts) and basic metadata.

• list_pdf_files(directory="all")

Human‑readable list of PDFs under allowed directories. directory filters by basename substring.

• show_accessible_directories()

JSON with accessible_directories, directory_count, max_file_size_mb, allowed_extensions.

Page range examples

• "first" → first page
• "last" → last page
• "3" → page 3
• "10-20" → pages 10 through 20 (inclusive)
• omit the argument → all pages

Quick start in Claude

Ask natural-language prompts like:

• “List the PDFs you can see under C:\Temp.” → list_pdf_files
• “What annotations are on page 162 of ‘Foo.pdf’?” → extract_annotations with page_range: "162"
• “Show me the highlighted text for page 162.” → extract_highlights_with_context with page_range: "162"
• “Read the first page text and metadata.” → read_pdf_text with page_range: "first"

Troubleshooting

• `ModuleNotFoundError: No module named 'pdf_memo'`

Ensure your repo root is on PYTHONPATH (see examples above) or run main.py from the repo root. This repo also injects sys.path in main.py for convenience.

• No highlights found

extract_highlights_with_context only returns Highlight annotations. If your file only has links/popups, use extract_annotations and adjust include_types.

• Jumbled text in highlights

The mapper prefers QuadPoints when present and falls back to Rect. It first tries within_bbox().extract_text() and then a (y,x) word grouping heuristic. Some PDFs with complex layouts may still need tuning.

License

MIT License

Copyright (c) 2025 Yong Choi

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.