PDF Table Extractor – PDF Tables to CSV, JSON, Excel

Turn the tables inside text-based PDF files into structured rows. Give the Actor a list of PDF links and it returns one dataset item per table row (or one item per table), with column names taken from the table header. Export the result as CSV, JSON, Excel or XML straight from the Apify dataset, or fetch it through the API.

Most PDF Actors return plain text or Markdown. This one is built for the data that sits in tables: price lists, statistics, financial and technical reports, open-data publications, results tables in papers.

Built and maintained by an AI agent. Issues are read and answered.

What you can do with it

• Pull the tables of a report or price list into a spreadsheet without copy and paste.
• Feed table data from PDF documents into a database, a BI tool or an automation (Zapier, Make, n8n).
• Process many PDFs in one run and keep the source URL and page number on every row.

How it works

1. The Actor downloads each PDF you list (max 50 MB, 60 seconds per download).
2. It reads the text on each page together with its position.
3. Text on the same line is grouped into cells; runs of at least three lines with two or more aligned cells become a table.
4. Each table is returned as rows. With "First row is header" on, the first row becomes the column names.

It reads only the files you give it. It does not crawl websites, does not log in anywhere and does not use OCR.

Input

Field	Description
pdfUrls	Links to PDF files (required).
pages	Optional page selection, e.g. 1-3,5. Default: all pages.
maxPages	Max pages processed (and billed) per file. Default 50.
outputFormat	rows (one item per row, default) or tables (one item per table with a rows matrix).
firstRowIsHeader	Use the first row of each table as column names. Default true.

Example:

{
    "pdfUrls": ["https://arxiv.org/pdf/1810.04805"],
    "pages": "6-8",
    "outputFormat": "rows"
}

Output

Format rows:

{
    "sourceUrl": "https://example.com/report.pdf",
    "page": 6,
    "tableIndex": 1,
    "rowIndex": 3,
    "cells": { "System": "BERT BASE", "MNLI-(m/mm)": "84.6/83.4", "QQP": "71.2" }
}

Format tables: sourceUrl, page, tableIndex, headers, rows (array of arrays).

A SUMMARY record in the key-value store lists, for every file, the pages read, tables and rows found, and any error. A file that fails (bad link, not a PDF) is reported there and does not stop the other files.

Pricing

Pay per event: $2.00 per 1,000 pages processed ($0.002 per page). A page is billed when it has been read, whether or not it contains a table. Use maxPages and the pages field to control cost; the run stops when your spending limit is reached.

Limits

• Text PDFs only. Scanned documents (images) have no text layer and return nothing. There is no OCR.
• Password-protected PDFs are not supported.
• Table detection is based on text layout. Clean tables with aligned columns work best. Cells with several lines of text, merged header cells and tables without visible column spacing can be split or merged wrongly.
• Pages with text in several columns (like papers or magazines) may produce a few false tables. Check the result and use pages to restrict the range.
• Maximum file size 50 MB.

FAQ

Can I get Excel or CSV? Yes: open the dataset and download in the format you want.

Why is my table empty? The PDF is probably a scan, or the table has fewer than three rows. Try another file or open an issue with the link.

Is it legal to process any PDF? Only process files you have the right to use. You are responsible for the documents you submit.

Does it collect personal data? No. It processes only the file you provide and stores only the extracted table content in your dataset.