MarkItDown Skill

Microsoft's Python utility for converting various file formats to Markdown for LLM and text analysis pipelines.

Overview

MarkItDown converts documents while preserving structure (headings, lists, tables, links). It's optimized for LLM consumption rather than human-readable output.

Supported Formats

Quick Start

Installation

# Full installation (recommended)
pip install 'markitdown[all]'

# Minimal with specific formats
pip install 'markitdown[pdf,docx,pptx]'

# Using uv
uv pip install 'markitdown[all]'

Optional Dependencies

Command-Line Usage

# Basic conversion
markitdown document.pdf > output.md

# Specify output file
markitdown document.pdf -o output.md

# Pipe input
cat document.pdf | markitdown > output.md

# With Azure Document Intelligence
markitdown document.pdf -o output.md -d -e "<endpoint>"

Python API

from markitdown import MarkItDown

# Basic conversion
md = MarkItDown()
result = md.convert("document.xlsx")
print(result.text_content)

# With LLM for image descriptions
from openai import OpenAI

client = OpenAI()
md = MarkItDown(
    llm_client=client,
    llm_model="gpt-4o",
    llm_prompt="Describe this image in detail"
)
result = md.convert("image.jpg")
print(result.text_content)

# With Azure Document Intelligence
md = MarkItDown(docintel_endpoint="<your-endpoint>")
result = md.convert("complex-document.pdf")
print(result.text_content)

Common Use Cases

Batch Convert Directory

from markitdown import MarkItDown
from pathlib import Path

md = MarkItDown()
input_dir = Path("./documents")
output_dir = Path("./markdown")
output_dir.mkdir(exist_ok=True)

for file in input_dir.glob("*"):
    if file.is_file():
        try:
            result = md.convert(str(file))
            output_file = output_dir / f"{file.stem}.md"
            output_file.write_text(result.text_content)
            print(f"Converted: {file.name}")
        except Exception as e:
            print(f"Failed: {file.name} - {e}")

Process for LLM Context

from markitdown import MarkItDown

def prepare_for_llm(file_path: str) -> str:
    """Convert document to LLM-ready markdown."""
    md = MarkItDown()
    result = md.convert(file_path)

    # Add source reference
    content = f"# Source: {file_path}\n\n{result.text_content}"
    return content

# Use with your LLM
context = prepare_for_llm("report.pdf")

Extract YouTube Transcript

# CLI
markitdown "https://www.youtube.com/watch?v=VIDEO_ID" > transcript.md

# Python
from markitdown import MarkItDown

md = MarkItDown()
result = md.convert("https://www.youtube.com/watch?v=VIDEO_ID")
print(result.text_content)

Image OCR with AI Description

from markitdown import MarkItDown
from openai import OpenAI

# Initialize with LLM support
client = OpenAI()
md = MarkItDown(
    llm_client=client,
    llm_model="gpt-4o"
)

# Convert image with AI description
result = md.convert("screenshot.png")
print(result.text_content)

Convert Jupyter Notebook

from markitdown import MarkItDown

md = MarkItDown()
result = md.convert("analysis.ipynb")
print(result.text_content)  # Code cells, outputs, markdown

Extract Wikipedia Content

from markitdown import MarkItDown

md = MarkItDown()
result = md.convert("https://en.wikipedia.org/wiki/Python")
print(result.text_content)  # Main article content only

Parse RSS Feed

from markitdown import MarkItDown

md = MarkItDown()
result = md.convert("https://example.com/feed.xml")
print(result.text_content)  # Feed entries as markdown

Plugin System

MarkItDown supports third-party plugins for extended functionality.

# List installed plugins
markitdown --list-plugins

# Enable plugins during conversion
markitdown --use-plugins document.pdf

# Enable plugins in Python
md = MarkItDown(enable_plugins=True)
result = md.convert("document.pdf")

Search GitHub for #markitdown-plugin to find available plugins.

MCP Server Integration

MarkItDown offers an MCP (Model Context Protocol) server for integration with LLM applications like Claude Desktop.

# Install MCP server
pip install markitdown-mcp

# Or from source
git clone https://github.com/microsoft/markitdown.git
cd markitdown/packages/markitdown-mcp
pip install -e .

See [markitdown-mcp][mcp-repo] for configuration details.

[mcp-repo]: https://github.com/microsoft/markitdown/tree/main/packages/markitdown-mcp

Docker Usage

# Build image
docker build -t markitdown:latest .

# Convert file
docker run --rm -i markitdown:latest < document.pdf > output.md

Troubleshooting

Common Errors

# ModuleNotFoundError for specific format
pip install 'markitdown[pdf]'  # Install missing dependency

# Azure authentication
export AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT="<endpoint>"
export AZURE_DOCUMENT_INTELLIGENCE_KEY="<key>"

Requirements

• Python >= 3.10
• Virtual environment recommended

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # Linux/macOS
.venv\Scripts\activate     # Windows

# Install
pip install 'markitdown[all]'

References

• references/cli-reference.md - Complete CLI options
• references/api-reference.md - Python API details
• references/examples.md - Extended examples
• references/advanced-features.md - Custom converters, URI handling
• GitHub: <https://github.com/microsoft/markitdown>;
• PyPI: <https://pypi.org/project/markitdown/>;

Gotchas

• DOCX with embedded images: images extract to separate files; markdown uses absolute paths — moving the markdown file alone breaks the image refs.
• PDF OCR confidence isn't surfaced — low-confidence text is returned as if certain; downstream LLM use can be confidently wrong.
• XLSX merged cells extract as separate cells with empty values for non-anchor positions — pivoted reports lose their column groupings invisibly.
• HTML to markdown loses CSS-driven layout — column-positioned tables collapse to row-major linear output; complex tables become unparseable.
• The --use-llm flag for image descriptions silently falls back to filename if no OPENAIAPIKEY — outputs look populated but contain no real description.