HTML Table to CSV

What this does

Paste the HTML of a web page, an email or a saved report, and this tool pulls out a table and writes it as CSV. It understands header cells (<th>) and data cells (<td>), decodes HTML entities such as & and  , handles merged cells, and produces output that follows RFC 4180, the CSV specification, so it opens correctly in Excel, Google Sheets, pandas and the usual libraries.

How it works

The tool does not need a browser DOM. It tokenizes the markup itself, ignores comments, <script> and <style> content, and collects the rows of the table you choose.

1. Which table. Tables are counted in document order starting at 1, counting only top-level tables. The result panel shows how many it found. A table nested inside a cell is not extracted separately; its text is flattened into the cell that contains it, with a space between its cells.
2. Cell text. Entities are decoded, runs of whitespace (including newlines from pretty-printed markup) collapse to a single space, and the text is trimmed. A <br> becomes a real line break inside the cell, which CSV can represent when the field is quoted.
3. Quoting. A field is wrapped in double quotes when it contains the delimiter, a double quote, a carriage return or a line feed, and any double quote inside it is doubled. Nothing else is quoted, which keeps the output readable.
4. Merged cells. A CSV has no merged cells, so the tool lays the table out on a grid the way a browser does. A cell with colspan="2" or rowspan="2" covers several grid positions. By default its text is repeated in each covered position, so every row is complete and sortable; choose blank mode to keep the text in the top-left position only. Each span is capped at 200, a table whose spans expand to more than 2 million grid cells is rejected with an error, and rowspan="0" is treated as 1.
5. Row and column counts. Rows are written in the order they appear in the source (thead, tbody and tfoot are all included in that order), and short rows are padded to the widest row.

Edge cases and limits

• Missing end tags are tolerated. An unclosed <td> or <tr> ends when the next cell or row begins, the way browsers recover.
• Only the entities in a built-in list are decoded by name, along with all numeric entities (—, —). A rare named entity is left as written.
• Tables built from <div> elements or CSS grids are not tables. Only real <table> markup is read.
• No JavaScript runs, so a table that a page fills in with scripts after load will be empty in the HTML source. Copy the rendered markup from your browser's developer tools (Inspect, then Copy outerHTML) instead of using View Source.
• Spreadsheet formula injection. A cell that starts with =, +, - or @ can be interpreted as a formula by Excel. This tool does not alter cell text, so review untrusted data before opening it in a spreadsheet.
• Line endings default to LF. Choose CRLF if the consumer insists on the exact form the RFC specifies. Input is limited to 2 MB.

When to use it

• Getting a data table out of a web page or HTML email into a spreadsheet.
• Converting an HTML report generated by a build tool, such as a coverage or test report, into something you can diff or chart.
• Rescuing a table from a CMS page where only the rendered HTML survives.

After conversion, the CSV to JSON tool turns the result into objects, and the CSV Diff tool compares it with an earlier export.