@llamaindex/liteparse-rest

An Express server that exposes @llamaindex/liteparse as an HTTP parsing backend. It supports single-file parsing, page screenshotting, and document complexity estimation.

For a full production setup with built-in rate limiting (Redis), distributed tracing (OpenTelemetry → Jaeger), and metrics (OpenTelemetry → Prometheus → Grafana), see the Docker Compose example.

Prefer gRPC? This repo also ships @llamaindex/liteparse-grpc — a gRPC server (and matching client) exposing the same LiteParse capabilities. See its README for details.

Table of contents

• Installation
• Quick start
• Docker
• Full server setup
• API specification
• Testing with the test script
• Related packages

Installation

npm install -g @llamaindex/liteparse-rest
# or
pnpm add -g @llamaindex/liteparse-rest

You can also invoke it directly without installing:

npx @llamaindex/liteparse-rest

Quick start

Once installed, launch the server with:

liteparse-rest-server

The server listens on port 5707 by default. Point a client at http://localhost:5707 and start hitting the API.

The published binary is the slim build — it removes built-in Redis caching, rate limiting and OpenTelemetry observability so it has no external dependencies. If you need those features, use the full server setup.

Docker

The repo ships a slim.Dockerfile that produces a self-contained image with no observability dependencies:

# Build the image
docker build -f slim.Dockerfile -t liteparse-rest .

# Run exposing port 5707
docker run -p 5707:5707 liteparse-rest

The API is then available at http://localhost:5707.

Full server setup

If you want an all-in-one deployment with built-in observability, caching, and rate-limiting, follow the guide in examples/docker-compose.

API specification

The server exposes three endpoints:

Method	Path	Description
POST	/parse	Parse a file into JSON pages, text, or markdown
POST	/screenshots	Render document pages as PNG images (NDJSON stream)
POST	/is-complex	Estimate document complexity / need for OCR

The full specification — request fields, query parameters, response shapes and error codes — is available in API_SPEC.md.

Testing with the test script

scripts/server-test.py is a self-contained script that uses uv to manage its own dependencies (httpx). Make sure the server is running before using it.

Usage

python scripts/server-test.py <command> <path> [options]

Or, if uv is available, run it directly:

./scripts/server-test.py <command> <path> [options]

Commands

Parse a single file (JSON pages response)

./scripts/server-test.py file path/to/document.pdf

Parse a single file (plain text or markdown response)

# plain text
./scripts/server-test.py file path/to/document.pdf text
# markdown
./scripts/server-test.py file path/to/document.pdf markdown

Screenshot pages of a document

# Screenshot all pages — images saved to the current directory
./scripts/server-test.py screen path/to/document.pdf

# Screenshot specific pages (comma-separated, 1-based)
./scripts/server-test.py screen path/to/document.pdf pages=1,2,3

Screenshots are saved as page_0.png, page_1.png, etc. in the current working directory.

Estimate document complexity

./scripts/server-test.py is-complex path/to/document.pdf

Related packages

This repository is a pnpm workspace containing:

Package	Description
@llamaindex/liteparse-rest	This package — an Express HTTP server wrapping LiteParse
@llamaindex/liteparse-grpc	A gRPC server + client for the same LiteParse capabilities, with a bundled .proto schema

License

MIT