HTML Table Extractor

Pull every data table from any web page into clean rows, as JSON, CSV or markdown.

Why tables break other extractors

Tables look simple and are not. Four things go wrong, and all four are handled here.

colspan and rowspan silently misalign everything. A cell spanning two rows shifts every later cell one column to the left, so row three onward quietly contains the wrong values. Nothing errors, the data is just wrong. This Actor expands spans into a proper rectangular grid, repeating spanned values so every row stands alone.

Stacked headers lose their meaning. Tables with grouped columns have two or three header rows, and taking only the first gives you a bare Revenue with no idea which year. This Actor joins them: 2025 / Revenue.

Navigation and layout tables come through as data. Wikipedia navboxes, infoboxes and sidebars are structurally identical to data tables. This Actor filters them by class, by role="presentation", and by link density, since a table that is more than 80% link text is a menu, not data. On one Wikipedia page that reduced six "tables" to the two that were real.

CSS ends up inside your cells. A <style> tag inside a table is picked up by naive text extraction, so cells fill with .mw-parser-output .navbar{display:inline...}. Removed here, along with footnote markers like [1].

What you get

Field	Description
url	Source page
tableIndex	Position of the table on the page
caption	Table caption, or the nearest heading above it
headers	Column names, with stacked headers joined
rows	Array of objects keyed by column name
rowCount / columnCount	Table dimensions
markdown	The table rendered as markdown, ready to feed an LLM
scrapedAt	ISO timestamp

Set One record per row to flatten the output so it exports straight to CSV or a spreadsheet.

Example input

{
  "urls": ["https://en.wikipedia.org/wiki/List_of_largest_companies_by_revenue"],
  "minRows": 3,
  "rowRecords": false
}

Options

• Minimum rows and Minimum columns — filter out small layout tables
• One record per row — flatten for CSV export
• Include markdown — a markdown rendering per table, useful for RAG, since plain-text extraction turns tables into unusable word salad
• Max tables per page — cap for pages with very many tables

Notes

• Reads the HTML the server returns. Tables rendered entirely by JavaScript after load will not appear.
• Nested tables are skipped rather than flattened into their parent cell.