Turn the documents a website publishes into Markdown, from Python or the command line: give a domain and get back every PDF, Word, Excel and CSV file linked from its pages and its sitemap, each converted to clean Markdown (headings, tables, lists, links) with its title, page count, word count and the page that links it. It works by domain, up to 50 documents per run, ready for an LLM, a RAG index or a search over a site's reports and forms. The code is free and open source (MIT). It runs on the QuanticData Data API, so there is nothing to download or parse on your side: sign in once and run a command.
pip install git+https://github.com/quantumproxies/pdf-to-markdown
pdf-to-markdown login
pdf-to-markdown documents bitcoin.org --doc-types pdf -o bitcoin_docs.jsonllogin opens the QuanticData sign-in page in your browser. New accounts get $2 of free usage every month, no card required: create one.
| Command | Input | Returns | Per run |
|---|---|---|---|
pdf-to-markdown documents |
domain or site URL | Markdown, title, kind, pages, words, where it is linked, 12 fields per row (fields and price) | up to 50 |
--doc-types pdf docx xlsx csv picks the formats (default PDF and Word). Each document is one row with its Markdown in markdown; -o docs.jsonl keeps one document per line, the usual input of a RAG loader.
A real row from pdf-to-markdown documents bitcoin.org --doc-types pdf --max-results 2 (2026-10-23):
{
"url": "https://bitcoin.org/bitcoin.pdf",
"kind": "pdf",
"chars": 21711,
"pages": 9,
"title": "Bitcoin: A Peer-to-Peer Electronic Cash System",
"words": 3656,
"author": null,
"source": "sitemap",
"found_on": null,
"markdown": "# Bitcoin: A Peer-to-Peer Electronic Cash System\n\n**Abstract.** A purely peer-to-peer version of electronic cash would allow online payments to be sent directly from one party to another without going through a financial institution. Digital signatures provide part of the solution, but the main benefits are lost if a trusted third party is still required to prevent double-spending.",
"truncated": false,
"scanned_pages": 0
}from quanticdata_pdf_to_markdown import PdfToMarkdown
pdfs = PdfToMarkdown() # saved login, or QUANTICDATA_API_KEY
docs = pdfs.documents("bitcoin.org", doc_types=["pdf"], max_results=10)
for d in docs:
print(d["kind"], d["pages"], d["words"], d["title"], d["url"])
print(docs.rows[0]["markdown"][:500])
docs.to_jsonl("bitcoin_docs.jsonl")Each call returns a Run: iterate it for the rows (plain dicts), take len(), or save it with to_csv(), to_json(), to_jsonl() or to_dataframe() (pandas). Errors are specific: InsufficientCredit, RateLimited, InvalidInput, JobFailed, AuthError.
- A site's documents as .md files: Save every PDF and Word file of a site as a Markdown file named after its URL, with an index.csv of titles, page counts and word counts. See
examples/save_markdown.py. - Chunks for a RAG index: Split each document's Markdown at its headings into chunks of bounded size, with source URL, title and section, one JSON line per chunk, ready for an embedding job. See
examples/rag_chunks.py. - Document inventory: List what several sites publish (type, title, pages, words, where each file is linked) in one CSV, with the Markdown kept short. See
examples/document_inventory.py.
Results print as a table; -o file.csv, -o file.json or -o file.jsonl writes a file. List inputs take a b c, a,b,c, @file.txt (one per line) or - (stdin). --no-wait starts a long run in the background and prints its id; pdf-to-markdown run <id> -o out.csv collects it later.
Every PDF and Office document a domain publishes, converted to clean LLM-ready Markdown.
pdf-to-markdown documents DOMAIN [options]
| Option | Type | Default | Description |
|---|---|---|---|
DOMAIN |
string | Domain or site section to collect documents from, e.g. bitcoin.org or www.irs.gov/forms-instructions. | |
--doc-types |
pdf / docx / xlsx / csv | Which document types to collect: pdf, docx, xlsx, csv (default pdf and docx). | |
--max-chars-per-doc |
integer | 40000 |
Cap on the Markdown returned per document, 1,000 to 200,000 characters (default 40,000); truncated says when a document reached it. |
--country |
string | Country to run the request from, as an ISO 3166-1 alpha-2 code (us, gb, de...). | |
--max-results |
integer | 10 |
How many documents to deliver at most (1–50). You pay only for delivered documents. |
pdf-to-markdown documents: 12 fields
| Field | Type | Description |
|---|---|---|
url |
string | Document URL. |
source |
string | Where the link was found: page | sitemap. |
kind |
string | pdf | docx | xlsx | csv. |
title |
string or null | Document title (PDF metadata or first heading). |
markdown |
string | The document converted to Markdown (headings, tables, lists, links). |
chars |
integer | Markdown length in characters. |
words |
integer | Word count of the Markdown. |
pages |
integer or null | Page count (PDF only). |
author |
string or null | Author from document metadata, when present. |
scanned_pages |
integer | Image-only pages, without a text layer. |
truncated |
boolean | True when the content hit max_chars_per_doc or the page cap. |
found_on |
string or null | The page of the site that links the document (null when it came from the sitemap). |
pdf-to-markdown loginsigns you in with OAuth 2.1 and PKCE in the browser. The token is stored in~/.config/quanticdata/credentials.json(readable only by you) and refreshed automatically. One login serves every QuanticData tool on the machine.- No browser on the machine (SSH, Docker, Colab): run
pdf-to-markdown login --no-browser, open the link anywhere, then paste back the address the browser ends on. - Scripts and CI: set
QUANTICDATA_API_KEYto a key from the dashboard, or store one withpdf-to-markdown login --api-key. pdf-to-markdown logoutrevokes the token. The connection is listed under API keys in the dashboard as "API · QuanticData CLI"; deleting that key disconnects every machine that uses it.
The code is free and MIT-licensed. Runs are billed to your QuanticData account per delivered row, at the price shown on each collector page linked above; $2 of usage is free every month and failed runs are never billed. Plans with higher limits: pricing.
The same data over plain HTTP:
curl -X POST https://api.quanticdata.io/v1/scraper/collectors/site_documents/run \
-H "Authorization: Bearer $QUANTICDATA_API_KEY" \
-H "Content-Type: application/json" \
-d '{"input": {"domain": "bitcoin.org", "doc_types": ["pdf"], "max_results": 2}}'- API reference with Python, Node, PHP and curl examples.
- MCP server to give Claude, Cursor or any MCP client the same tools.
- n8n node for no-code workflows.
Is this PDF to Markdown scraper free?
The code is free and MIT-licensed. Scraping runs on your QuanticData account: $2 of usage is free every month with no card, then you pay per delivered row at the price shown on each collector page. Failed runs are never billed.
Which file types does it read?
PDF, Word (.docx), Excel (.xlsx) and CSV, chosen with --doc-types; the default is PDF and Word.
How does it find the documents?
From the links on the site's pages and from its sitemap. Each row says which (source) and, for a linked file, the page that links it (found_on).
Are tables and headings kept?
Yes. Headings, tables, lists and links are converted to Markdown, and the row adds the document title, page count, word count and author when the file states one.
How much can one run return?
Up to 50 documents per run and up to 200,000 characters of Markdown per document (--max-chars-per-doc, default 40,000); ask for fewer documents with --max-results and you pay only for what is delivered.
Can I export the Markdown?
Add -o docs.jsonl for one document per line, or -o docs.csv with the Markdown in one column; examples/save_markdown.py writes one .md file per document.
- Web Scraper API: Web scraper API for Python and the command line.
- llms.txt and AI Crawler Checker: llms-txt checker and AI crawler checker for Python and the command line.
- Google SERP API: Google SERP API for Python and the command line.
MIT. Built and maintained by QuanticData.