extracting-tables

Extracting tables

Use this when the user wants tabular data out of HTML — pricing tables, data grids, spec sheets. html-to-markdown parses <table> elements into two surfaces at once: inline GFM Markdown tables in the content stream, and a structured tables array in the JSON output.

Two surfaces

# Inline GFM tables appear in the Markdown body
html-to-markdown input.html

# Structured table data appears under result.tables (JSON)
html-to-markdown --json input.html | jq '.tables'

• Markdown tables in content — | col | col | blocks, good for LLM ingestion and human reading.
• Structured tables array — each entry has a markdown field (pre-rendered) and a grid of structured cells (rows × cols). Use this when downstream code needs exact cell access.

Both are populated from the same parse; you do not need a flag to enable table parsing.

Extraction-only

When you only care about tables (not the Markdown body):

html-to-markdown --json --no-content input.html | jq '.tables'

Inspecting tables

# How many tables, and the row count of each
html-to-markdown --json input.html \
  | jq '.tables | to_entries | map({index: .key, rows: .value.grid.rows, cols: .value.grid.cols})'

# Just the rendered markdown of the first table
html-to-markdown --json input.html | jq -r '.tables[0].markdown'

Line breaks in cells

By default <br> inside a cell is converted to a space. Keep hard breaks:

html-to-markdown input.html --br-in-tables

Programmatic access

from html_to_markdown import convert, ConversionOptions

html = "<table><tr><th>A</th><th>B</th></tr><tr><td>1</td><td>2</td></tr></table>"
# result.tables is only populated when include_document_structure=True
result = convert(html, ConversionOptions(include_document_structure=True))
for table in result.tables:
    print(table.markdown)            # rendered GFM markdown
    print(table.grid.cells[0].content)  # first cell (grid is a TableGrid)