markitdown-plus 0.2.0

Batch conversion, asset extraction, and RAG-ready output toolkit for Microsoft MarkItDown.

MarkItDown Plus

Batch conversion, asset extraction, RAG-ready Markdown, JSONL chunks, and cleaner AI document pipelines for Microsoft MarkItDown.

MarkItDown Plus is an enhancement toolkit built on top of Microsoft MarkItDown. It adds folder conversion, recursive processing, optional parallel workers, Markdown cleanup, multiple chunking strategies, lightweight asset extraction, conversion manifests, and JSONL output for RAG workflows.

This project is independent and is not affiliated with Microsoft. It is designed as a companion CLI for the Microsoft MarkItDown ecosystem.

Why MarkItDown Plus?

Microsoft MarkItDown is excellent for converting individual files to Markdown. MarkItDown Plus focuses on the next step: turning many documents into clean, AI-ready project output.

Key features:

• Batch convert files and folders
• Recursive directory conversion
• Parallel conversion with --workers
• Optional tqdm progress with --progress
• RAG-ready JSONL chunk export
• Chunk strategies: heading, fixed, semantic-lite
• Markdown cleanup for common PDF/document artifacts
• Basic asset extraction for DOCX / PPTX / XLSX / HTML
• manifest.json, failed.json, and large-run JSONL manifest streaming
• Unicode-safe output filenames
• PayPal funding link included through GitHub Sponsors/Funding

Installation

pip install markitdown-plus

For progress bars:

pip install "markitdown-plus[progress]"

For development tests and coverage:

pip install -e ".[dev]"
pytest

Quick Start

Convert a folder:

markitdown-plus convert ./docs --output ./out

Convert recursively:

markitdown-plus convert ./docs --output ./out --recursive

Convert only specific file types:

markitdown-plus convert ./docs --output ./out --types pdf,docx,pptx,xlsx,html,csv

Clean Markdown and export RAG chunks:

markitdown-plus convert ./docs --output ./out --clean --rag

Use parallel workers:

markitdown-plus convert ./docs --output ./out --recursive --workers 4 --progress

Use auto worker count:

markitdown-plus convert ./docs --output ./out --workers 0

Extract assets when supported:

markitdown-plus convert ./docs --output ./out --extract-assets

Use a specific chunking strategy:

markitdown-plus convert ./docs --output ./out --rag --chunk-strategy semantic-lite

Output Structure

A normal batch run creates:

out/
  markdown/
    report.md
  metadata/
    report.json
  manifest.json

With RAG enabled:

out/
  markdown/
    report.md
  chunks/
    report.jsonl
  metadata/
    report.json
  manifest.json

With asset extraction enabled:

out/
  markdown/
    report.md
  assets/
    report_img_001.png
    report_img_002.jpg
  metadata/
    report.json
  manifest.json

For very large jobs, MarkItDown Plus avoids huge manifest.json files by streaming records:

out/
  manifest.json
  manifest-records.jsonl
  failed.jsonl

Chunk Strategies

heading

Default. Preserves Markdown heading paths and is best for most structured documents.

markitdown-plus convert ./docs -o ./out --rag --chunk-strategy heading

fixed

Creates stable chunk sizes and ignores heading boundaries. Useful for embedding pipelines that prefer consistent lengths.

markitdown-plus convert ./docs -o ./out --rag --chunk-strategy fixed

semantic-lite

Dependency-free rule-based topical splitting. It starts new chunks at obvious semantic cues such as headings, summary, conclusion, recommendations, and other section-like paragraphs.

markitdown-plus convert ./docs -o ./out --rag --chunk-strategy semantic-lite

Asset Extraction

--extract-assets currently supports lightweight extraction for:

• .docx
• .pptx
• .xlsx
• .html / .htm local image references

PDF image extraction is intentionally left for a later version because reliable PDF asset extraction requires heavier format-specific dependencies.

When assets are extracted, MarkItDown Plus appends an Extracted Assets section to the generated Markdown and records asset metadata in the file-level metadata JSON.

Single File Commands

Convert one file directly:

markitdown-plus single report.pdf -o report.md

Clean an existing Markdown file:

markitdown-plus clean dirty.md -o clean.md

Chunk an existing Markdown file:

markitdown-plus chunk clean.md -o chunks.jsonl --chunk-strategy fixed

Development

git clone https://github.com/lamguo/markitdown-plus.git
cd markitdown-plus
pip install -e ".[dev]"
pytest

The test configuration includes a coverage gate:

pytest --cov=markitdown_plus --cov-fail-under=85

Optional property and benchmark tests are included. They are skipped automatically if hypothesis or pytest-benchmark is not installed.

GitHub Topics

Suggested topics for the repository:

markitdown
microsoft-markitdown
markdown
rag
llm
document-conversion
pdf-to-markdown
docx-to-markdown
batch-conversion
jsonl
asset-extraction
ai-tools

Support This Project

If MarkItDown Plus helps you save time or build better AI document pipelines, you can support development here:

• Star this repository
• Support via PayPal: https://www.paypal.me/lamguo

Thank you for supporting open-source development.

License

MIT License.