liteparse

name	liteparse
description	Fast document parsing for RAG pipelines via LiteParse (Rust, 100x faster than Python parsers)
tags	["rag","document-parsing","rust","pdf","llm"]
related_skills	["crawl4ai","huggingface-hub"]

LiteParse — Fast Document Parsing for RAG

Parse PDFs, DOCX, HTML, and other documents 100x faster than Python parsers. From the LlamaIndex team.

Prerequisites

• Rust toolchain (or Python pip)
• LiteParse installed: cargo install liteparse or pip install liteparse

Usage

CLI — Parse a single document

liteparse document.pdf --output ./parsed/

CLI — Parse a directory

liteparse ./documents/ --recursive --output ./parsed/

Python API — Integration with RAG pipelines

from liteparse import parse

result = parse("document.pdf")
print(result.markdown)   # Clean markdown output
print(result.tables)     # Extracted tables as dataframes
print(result.images)     # Extracted images as paths
print(result.metadata)   # Title, author, dates

MCP Server — For AI agents

liteparse serve --port 8091
# Add to agent config:
# "mcpServers": {"liteparse": {"command": "liteparse", "args": ["serve", "--port", "8091"]}}

Supported Formats

Format	Support	Notes
PDF	✅ Full	Text, tables, images, metadata
DOCX	✅ Full	Headings, lists, tables
HTML	✅ Full	DOM-aware extraction
Markdown	✅ Passthrough	Clean formatting
EPUB	✅ Full	Chapter structure preserved

Pitfalls

1. Scanned PDFs: LiteParse extracts embedded text. For scanned PDFs, pair with OCR (tesseract) first.
2. Complex tables: Multi-row spanning tables may need manual cleanup in output.
3. Memory usage: Large documents (>100MB) may need chunked processing.
4. Encoding: Non-UTF-8 documents may need preprocessing.

Verification

# Parse a test document and verify output
liteparse test.pdf --output ./test-output/
cat ./test-output/test.md  # Should contain extracted text
ls ./test-output/           # Should contain images if present