# PDFMD

PDFMD converts PDFs into Markdown with page and table metadata. Operated by Dark Vector Cognition.
English printed scans can be processed separately with OCR into a searchable PDF or Markdown.

- Website and full Markdown download: https://pdfmd.dev/
- API and connection instructions: https://pdfmd.dev/docs#agents
- Stateless remote MCP: https://pdfmd.dev/mcp
- English scanned PDF OCR: https://pdfmd.dev/ocr
- OCR REST API: https://pdfmd.dev/docs#ocr
- Privacy and cloud processing: https://pdfmd.dev/privacy
- Terms: https://pdfmd.dev/terms

MCP tool: convert_pdf_url accepts one public HTTP(S) PDF URL. It returns at most 20,000 Markdown characters, page counts, source URL and conversion date. Longer results are explicitly marked as excerpts; the full conversion uses the API page allowance. No result is saved or assigned a download URL.

Anonymous API allowance is 25 pages per UTC day per network address, at most 25 pages per PDF; a free key issued for an email at https://pdfmd.dev/billing/free allows 50 pages a day, at most 25 per PDF. Free use is for one person: your own projects or your studies. Production, team and commercial use need a paid plan. A public PDF may be up to 50 MB; the API result must fit 4 MB. Existing API keys may be supplied as an Authorization Bearer or x-api-key header. No account or key is needed for the free tier. Scanned pages are identified but are not OCR'd by this Markdown connector. Local file paths, private links and credentialed links are unsupported by this connector. The source must be one the caller is entitled to process.

POST /api/v1/ocr accepts a multipart file with an output field, or a JSON body containing url and output. Output is markdown (default) or searchable_pdf. English printed text only; handwriting and other languages are not supported at launch. OCR shares the existing API allowance at one page credit per page, including pages that already have text. The same input and page caps apply; returned JSON or PDF must fit 4 MB. Searchable PDF output returns application/pdf bytes with X-Pdfmd-Pages and X-Pdfmd-Ocr-Pages. Markdown output returns document JSON with operation=ocr, output=markdown, language=eng and ocrPages. Both outputs have X-Pdfmd-Rx: ocr. Error responses are JSON. Requests have up to five minutes; check /api/v1/me before retrying an interrupted request. OCR availability depends on the deployment; an unconfigured deployment returns 503 ocr_unavailable.

The browser OCR tool and the ordinary web converter share 5 files and 50 pages per UTC day per network address, at most 25 pages per PDF. Ordinary text-to-Markdown conversion holds files and results in memory. OCR uses private temporary worker files that application code removes on success, failure and worker timeout. No result has a saved public download URL. Hosting provider buffering and log retention have not been measured.

The connector uses the same page reservations, limits and public-network URL safeguards as the REST API. Calls record bounded outcomes and timing categories, never source links or document text. DNT and Global Privacy Control suppress measurement. Calls are not unique people, installs or purchases.