sergiudanstan/mcp-pdf-to-csv

PDF to CSV MCP Server

FastMCP server for extracting tables from PDF files and converting them to CSV format.

Features

• Batch PDF processing: Extract tables from single or multiple PDFs
• Multiple extraction strategies: Auto, lattice (grid-based), and stream (whitespace-based)
• Flexible output: Save individual tables or merge all tables per PDF
• Configurable folders: Set custom input/output directories

Installation

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install fastmcp pdfplumber pandas

Configuration

The server is configured in Claude Desktop at: ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "pdf-to-csv": {
      "command": "/Users/sara/MCP_PDF_CSV/.venv/bin/python3",
      "args": ["/Users/sara/MCP_PDF_CSV/pdf_to_csv_server.py"]
    }
  }
}

Usage

Default Folders

• Input: /Users/sara/MCP_PDF_CSV/pdfs/
• Output: /Users/sara/MCP_PDF_CSV/output_csv/

Available MCP Tools

1. set_folder(folder: str)

   • Set the input folder containing PDF files
   • Example: set_folder("/path/to/pdfs")

2. set_output(folder: str)

   • Set the output folder for CSV files
   • Example: set_output("/path/to/output")

3. list_pdfs()

   • List all PDF files in the current input folder

4. extract_tables(filename: str, pages: str = "all", strategy: str = "auto", merge: bool = False, strip_ws: bool = True)

   • Extract tables from a single PDF
   • Parameters:
     • filename: PDF filename in the input folder
     • pages: "all" or specific pages like "1,3-5"
     • strategy: "auto", "lattice", or "stream"
     • merge: Create merged CSV with all tables
     • strip_ws: Strip whitespace from cells

5. extract_all(pages: str = "all", strategy: str = "auto", merge: bool = False, strip_ws: bool = True)

   • Extract tables from all PDFs in the folder
   • Same parameters as extract_tables (except filename)

6. show_config()

   • Display current input/output folder configuration

Testing

Run the test script to verify installation:

source .venv/bin/activate
python test_server.py

Troubleshooting

Check Claude Desktop Logs

tail -50 ~/Library/Logs/Claude/mcp-server-pdf-to-csv.log

Common Issues

1. "Read-only file system" error: Fixed by using absolute paths (SCRIPT_DIR)
2. Connection errors: Restart Claude Desktop after configuration changes
3. Missing dependencies: Reinstall with pip install fastmcp pdfplumber pandas

Technical Details

• Framework: FastMCP 2.13.0.2
• PDF Library: pdfplumber (with pdfminer.six backend)
• Data Processing: pandas
• Transport: STDIO (Standard Input/Output)
• Protocol: MCP (Model Context Protocol) 2025-06-18

Files

• pdf_to_csv_server.py - Main MCP server
• test_server.py - Validation script
• README.md - This file
• .venv/ - Python virtual environment
• pdfs/ - Default input folder
• output_csv/ - Default output folder