The hardest part of PDF conversion is usually not the text — it's the tables. Copy-paste from a PDF into a text editor and your tables become garbled columns with no structure. mdstill reconstructs tables into proper GFM Markdown with pipes and alignment so downstream tools (ChatGPT, Claude, Obsidian, spreadsheet importers) can actually parse them.
Max file size: 20MB ·
A clean result means every row is a distinct Markdown line, columns are separated by pipes, and the header row is marked with a separator. Tools that understand GFM — GitHub itself, Obsidian, LLMs, most static-site generators — will render the table with proper alignment and can reason about it as structured data.
Tables with a clear visual grid, consistent column count across rows, and explicit header rows convert cleanly. Financial tables, comparison tables, reference tables, and most scientific result tables fall in this bucket. If the source PDF has selectable text and the table looks like a table to you when you open the PDF, there's a good chance the conversion lands correctly.
Tables with heavily merged cells, nested sub-headers, multi-line cells, or unusual layouts (like tables that span page breaks with repeated headers) are the failure modes. You'll usually still get something readable, but you may need a 30-second manual clean-up. For extremely complex tables, no general-purpose tool will beat a manual extract — this is true across all tools, including the ML-heavy ones.
The Markdown table is plain text, so you can pipe it anywhere: paste into a GitHub issue, drop into Obsidian, import into a Jupyter notebook, feed to an LLM for analysis. If you want a spreadsheet back, most tools (including Pandas via read_html after a tiny wrap) can round-trip a GFM table into rows and columns.