everything-markdown

🌐 English · 简体中文

Convert PDF, Word, HTML, URLs, PowerPoint, Excel, CSV, TSV, EPUB, RTF and ODT into clean, LLM-ready Markdown — with academic-paper-grade PDF quality.

pip install everything-markdown
everything-md paper.pdf -o paper.md

Why everything-markdown?

	everything-markdown	markitdown	docling	marker
Academic PDF quality (header/footer strip, caption reattach, line-number removal)	Yes	No	Partial	Partial
Formats supported	11	8	3	1
Install size	~150 MB	~50 MB	~2 GB	~1.5 GB
GPU required	No	No	Optional	Optional
MCP server	Yes	No	No	No
Agent skill (SKILL.md)	Yes	No	No	No
Scanned PDF OCR	Yes¹	No	Yes	Yes
LangChain loader	Yes	No	No	No

¹ Requires ocrmypdf + Tesseract. Auto-detected at runtime.

See docs/benchmarks.md for a detailed comparison.

Quick start

Python API

from everything_markdown import convert

doc = convert("paper.pdf")
print(doc.markdown)       # the full markdown text
print(doc.title)          # extracted document title
print(doc.warnings)       # any conversion warnings
print(doc.assets)         # extracted images (Asset objects)

# Write to file with images extracted:
doc = convert("paper.pdf", output_path="paper.md", images="folder")

CLI

# Basic conversion
everything-md paper.pdf                        # → paper.md
everything-md report.docx -o output.md         # explicit output path
everything-md https://example.com/post --stdout # stream to stdout

# Image handling
everything-md paper.pdf --images folder         # extract images (default)
everything-md paper.pdf --images base64         # inline as data URIs
everything-md paper.pdf --images skip           # drop images

MCP server (Claude Desktop / Cursor / any MCP client)

pip install everything-markdown[mcp]

Add to your MCP client config:

{
  "mcpServers": {
    "everything-markdown": {
      "command": "everything-md-mcp"
    }
  }
}

Or use uvx (no install needed):

{
  "mcpServers": {
    "everything-markdown": {
      "command": "uvx",
      "args": ["--from", "everything-markdown[mcp]", "everything-md-mcp"]
    }
  }
}

LangChain integration

pip install everything-markdown[langchain]

from everything_markdown.integrations.langchain import EverythingMarkdownLoader

loader = EverythingMarkdownLoader("paper.pdf")
docs = loader.load()
# docs[0].page_content  → full markdown
# docs[0].metadata      → {"source": "paper.pdf", "title": "...", ...}

Agent skill (Claude Code / Cursor / Codex CLI)

No pip install needed — the skill bundles its own code and resolves dependencies on first run.

# One-liner install:
curl -fsSL https://raw.githubusercontent.com/JustinRaoV/everything-markdown/main/install.sh | bash

# Or via Claude Code marketplace:
/plugin marketplace add JustinRaoV/everything-markdown
/plugin install convert-to-markdown@everything-markdown

Then tell your agent: 把 paper.pdf 转成 markdown

Install

Option 1: pip (recommended)

pip install everything-markdown           # core conversion
pip install everything-markdown[mcp]      # + MCP server
pip install everything-markdown[langchain]# + LangChain loader
pip install everything-markdown[ocr]      # + OCR for scanned PDFs
pip install everything-markdown[all]      # everything

Option 2: Agent skill (no pip)

Pick whichever path matches your agent:

# Zip download (no git, no shell):
# Grab from https://github.com/JustinRaoV/everything-markdown/releases/latest
unzip convert-to-markdown-v*.zip -d ~/.claude/skills/

# One-liner:
curl -fsSL https://raw.githubusercontent.com/JustinRaoV/everything-markdown/main/install.sh | bash

# Manual:
git clone https://github.com/JustinRaoV/everything-markdown.git
# Point your agent at: <clone>/skills/convert-to-markdown/

Runtime requirements

• Python 3.11+ must be on PATH.
• For agent skill mode: uv (recommended) or plain python3 (bootstrap falls back to venv + pip).
• For OCR: pip install everything-markdown[ocr] or install ocrmypdf + Tesseract system-wide.

Supported formats

Format	Extensions	Notes
PDF (text)	.pdf	pymupdf4llm + custom post-processor; paper-grade on academic PDFs; password-protected PDFs detected
PDF (scanned)	.pdf	Auto-OCR via ocrmypdf when available; falls back to warning if not installed
HTML / URL	.html, .htm, http(s)://	Readability + html2text; title from og/twitter/title/h1; charset meta respected
DOCX	.docx	Mammoth (no legacy .doc); title from docProps/core.xml; tables as GFM
PPTX	.pptx	python-pptx; bold/italic preserved; hyperlinks extracted; nested bullets; notes as blockquote
XLSX / CSV / TSV	.xlsx, .csv, .tsv	One GFM table per sheet; CSV/TSV auto-detects delimiter, BOM + CJK/Western encodings
EPUB	.epub	ebooklib; cross-chapter image dedup; all three image modes supported
RTF	.rtf	striprtf (pure Python); plain-text extraction; paragraphs + bullets preserved
ODT	.odt	odfpy (pure Python); headings/lists/tables/images; bold/italic, hyperlinks, footnotes/endnotes

What you get

$ everything-md ~/Downloads/whitepaper.pdf
wrote: /Users/.../Downloads/whitepaper.md
  + 12 asset(s) -> whitepaper.assets

The output has paragraphs reflowed across 2-column layouts, headings inferred from font hierarchy, GFM tables, and image refs pointing at the sibling assets folder.

For academic PDFs, the post-processor strips bioRxiv line numbers, CrossMark badges, license boilerplate, and page headers/footers — then reattaches figure captions to their images.

See examples/ for input/output pairs you can read straight on GitHub.

Development

git clone https://github.com/JustinRaoV/everything-markdown.git
cd everything-markdown
uv sync                                  # installs all deps
uv run pytest                            # 307 tests
uv run ruff check .                      # lint
make check                               # lint + tests + format check

Architecture: docs/technical-design.md. Adding a converter: CONTRIBUTING.md.

Troubleshooting

• First run takes 30+ seconds — expected; wheels download once.
• Scanned PDF produces empty markdown — install ocrmypdf + Tesseract, or pip install everything-markdown[ocr].
• Agent skill doesn't appear — restart your agent; confirm ~/.claude/skills/convert-to-markdown/SKILL.md exists.

Full list: docs/troubleshooting.md.

FAQ

• Can I use this in a non-Python project? Yes — the CLI is a standalone command, and the MCP server works with any MCP client.
• Where does the markdown get written? Alongside the input by default (paper.pdf → paper.md); pass -o to override, or --stdout to stream.
• Can I batch-convert a directory? Loop in your shell; the converter is intentionally one-document-per-invocation.

Full list: docs/faq.md.

Changelog

See CHANGELOG.md.

License

MIT