What it does
pdffr converts PDF files to Markdown in the browser or in Node. It reads each glyph's coordinates, size and font to rebuild headings, paragraphs, nested lists, tables, reading order and math written as LaTeX. Scans, stamps and images with burned-in text are sent to OCR, while native text is not rasterised. The MCP server exposes pdf_to_markdown, pdf_outline and pdf_tables tools, and LangChain.js, LlamaIndex.TS and Python loaders are also provided.
It suits agents or pipelines that need PDF content as Markdown without uploading documents to a cloud service. It is aimed at born-digital PDFs, where plain text extraction loses structure, and at mixed files that combine native text with scanned pages. A browser playground and a benchmark against other open-source PDF tools are included in the repository.
The MCP server is set up in clients such as Claude Desktop, Claude Code or Cursor. In the browser, pdf.js and tesseract.js run in web workers within the user's tab. The Node entry uses the same pipeline with @napi-rs/canvas as an optional peer, and pdfjs-dist and tesseract.js are peer dependencies. The Python package drives the command-line tool and needs Node 20 or later.
Limitations and requirements
The publisher describes the project as early and says it is accurate on the documents it was built against. Node usage needs the pdfjs-dist and tesseract.js peer dependencies, and @napi-rs/canvas is optional. The Python package requires Node 20 or later.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
pdffr
Local, private PDF to Markdown for agents: geometry-first parsing, OCR only where needed.
v0.1.1As shown on registry.modelcontextprotocol.io when we last read it