Document to Markdown Skills for Claude Code: PDF, Word and MarkItDown
The best skill for turning local documents into Markdown in Claude Code is firecrawl-parse (51,693 installs), which converts PDF, Word, Excel and HTML files to Markdown on disk. Firecrawl's anydoc (13,751 installs) needs no install, and two MarkItDown skills add audio, images and YouTube.
By Skillselion, an Ellelion LLC publication · Updated November 9, 2026 · 4 min read · Stats verified against the live catalog
Claude Code reads text well and binary office files badly, so a 40-page PDF or a quarterly spreadsheet needs converting before the agent can work with it. A document-to-Markdown skill does that conversion in one command, and firecrawl-parse from firecrawl/cli (51,693 installs) is the most-installed skill built only for turning local files into Markdown.
Key takeaways
- firecrawl-parse (51,693 installs) converts PDF, DOCX, DOC, ODT, RTF, XLSX, XLS and HTML into Markdown on disk, and can also summarize a file or answer a question about it.
- Firecrawl's anydoc skill (13,751 installs, 23,609 GitHub stars) runs through
npxwith no install and covers Word, PowerPoint, Excel, OpenDocument, EPUB, CSV and PDF. - Two catalog skills wrap Microsoft's MarkItDown: markitdown from davila7/claude-code-templates (2,600 installs) and K-Dense's markitdown (2,345 installs). Both cover audio transcription and YouTube URLs. Only davila7's lists image OCR; K-Dense's skill notes that MarkItDown does no local OCR on scanned pages and lists the markitdown-ocr plugin or Azure as options.
- Anthropic's pdf skill (232,199 installs, 197,250 GitHub stars) is the broadest PDF toolkit: text and tables with pdfplumber, merging with pypdf, OCR with pytesseract.
- For scanned papers with tables and formulas, the MinerU-based pdf-converter (9,718 installs) handles OCR in 80 languages through MinerU's hosted API. Web pages are a separate job, covered in the web scraping skills guide.
Why convert documents to Markdown before Claude Code reads them?
Markdown keeps the structure an agent needs (headings, lists, tables, links) and drops the layout bytes it does not. A converted file is also a normal text file, so the agent can grep it, read one section at a time and diff two versions. firecrawl-parse's own instructions tell the agent to always write output to a .firecrawl/ folder with -o, because a parsed document can run to hundreds of KB and fill the context window if streamed to the terminal. That one rule is the difference between reading a report and losing the session to it.
What is the best document to Markdown skill for Claude Code?
firecrawl-parse from firecrawl/cli is the best general pick, with 51,693 installs and 690 GitHub stars (skills.sh registry, GitHub). Install it with:
npx skills add https://github.com/firecrawl/cli --skill firecrawl-parse

Use it when a local path is in the prompt: firecrawl parse ./paper.pdf -o .firecrawl/paper.md for the full text, -S for an AI summary, or a question flag to query the file. The skill tells the agent to use scrape instead when the source is a URL. Parsing uses Firecrawl credits, about 1 per PDF page, and accepts files up to 50 MB.
If you do not want an account or a CLI install, convert-documents-to-markdown from firecrawl/anydoc (13,751 installs) runs npx -y @firecrawl/anydoc <file> -o out.md on Node 20 or later. It detects the format from file content and returns exit code 1 when a file cannot be converted. Scanned, image-only pages exit with code 3, because anydoc does no OCR itself. Rerun with --ocr hosted to send the file to Firecrawl Parse, which needs no signup.
Is MarkItDown better than other converters for Claude Code?
MarkItDown is the right choice when your inputs go beyond office files. The MarkItDown project from Microsoft converts PDF, DOCX, PPTX and XLSX, and also images with OCR, audio with transcription, ZIP archives, EPUB and YouTube URLs. davila7's markitdown skill (2,600 installs, 33,979 GitHub stars) wraps it for Claude Code with example scripts and an optional OpenRouter key for LLM image descriptions, and K-Dense's markitdown (2,345 installs, 57,724 GitHub stars) packages it for research workflows. For batch jobs, office-to-md from claude-office-skills (5,916 installs) uses the same library and adds archival patterns with metadata timestamps.
How do you get tables and scanned pages out of a PDF?
Use Anthropic's pdf skill for born-digital PDFs and a MinerU or OCR skill for scans. The pdf skill from anthropics/skills (232,199 installs) maps each task to a tool: pdfplumber for layout-aware text and tables into pandas, pypdf for merge, split and rotate, pdftotext and qpdf on the command line, and pytesseract with pdf2image for scanned pages. Its sibling docx skill (221,095 installs) does the same for Word files, including tracked changes.
For academic papers full of formulas, the MinerU pdf-converter (9,718 installs) outputs Markdown, Word, HTML, LaTeX or plain text with OCR in 80 languages. It uploads each file to MinerU's API. With no token, its flash mode caps a document at 10 MB and 20 pages and returns Markdown only; Word, HTML and LaTeX output need a token. smart-ocr from claude-office-skills (8,881 installs) covers images and photographed pages. If the end goal is a slide deck rather than text, the presentation skills guide picks up from there.
A complete starter stack
- firecrawl-parse (51,693 installs): local PDF, Word and Excel files to Markdown on disk.
- convert-documents-to-markdown from anydoc (13,751 installs): zero-install conversion through
npx. - markitdown (2,600 installs): audio, images, ZIP and YouTube on top of office formats.
- pdf from anthropics/skills (232,199 installs): tables, merging, forms and OCR for PDFs.
- defuddle (0 installs): web pages only, for the URL side of the job, since the other four handle files on disk.
Common pitfalls
- Streaming a converted file straight into the chat. firecrawl-parse and anydoc both tell the agent to write to a file with
-o, since one long PDF can run to hundreds of KB of text. - Sending scanned PDFs to a converter without OCR. anydoc exits with code 3 on image-only pages unless you opt in to
--ocr hosted, so route scans to the MinerU converter, Anthropic's pdf skill or that hosted option. - Installing a URL tool for local files. defuddle (0 installs) is built for web pages, while firecrawl-parse is built for files on disk.
Pick one converter for local files, one for scans, and browse the best docs and planning skills for the rest of the document workflow.
Common questions
What is the best skill to convert PDFs to Markdown in Claude Code?
firecrawl-parse from firecrawl/cli (51,693 installs) is the most-installed skill built for local files to Markdown, and Anthropic's pdf skill (232,199 installs) adds tables, merging and OCR. See firecrawl-parse.
Is there a MarkItDown skill for Claude Code?
Yes. davila7's markitdown (2,600 installs) and K-Dense's markitdown (2,345 installs) both wrap Microsoft's MarkItDown converter. See markitdown from davila7/claude-code-templates.
How do I convert a Word document to Markdown with Claude Code?
Install anydoc's convert-documents-to-markdown (13,751 installs) and run npx -y @firecrawl/anydoc report.docx -o report.md. See convert-documents-to-markdown.
Can Claude Code read scanned PDFs?
Yes, with an OCR step. The MinerU pdf-converter (9,718 installs) handles scans in 80 languages through MinerU's hosted API, and Anthropic's pdf skill uses pytesseract. See the MinerU pdf-converter.
What is the difference between defuddle and firecrawl-parse?
defuddle (0 installs) cleans web pages from a URL, while firecrawl-parse (51,693 installs) converts files on disk. See defuddle from kepano/obsidian-skills.
Ranked by Skillselion - an independent directory of AI-coding tools, not affiliated with Anthropic, OpenAI or Cursor. Tool rankings reflect real adoption (installs, then GitHub stars) from the skills.sh registry and GitHub, last updated November 9, 2026.