# Unstract

> No-code platform that pulls data out of documents with LLMs

Steal this: Describe in plain language what to extract and get clean JSON back from an API.

- Category: Agents & dev (`agents`)
- Tags: `open-source`, `no-code`, `mcp`
- Official: https://unstract.com
- GitHub: https://github.com/Zipstack/unstract
- Wunderlust: https://wunderlust.xyz/r/unstract
- Image: https://wunderlust.xyz/media/unstract-poster.jpg
- Video preview: https://wunderlust.xyz/media/unstract.webm
- Added: 2026-11-10

## What it is

A platform that uses LLMs to turn PDFs, images and scans into structured JSON. You describe what to extract in plain language in Prompt Studio, then publish it as an API or an ETL pipeline.

## What it’s for

For pulling data from invoices, forms and other documents without writing regex per layout.

## Benefits

- Schema defined once by prompt
- Publish as a REST API or ETL pipeline
- MCP server and an n8n node

## Caveats

- AGPL-3.0 license, with a separate paid enterprise edition
- Runs with Docker and needs 8 GB of RAM

## Similar on Wunderlust

- [OpnForm](https://wunderlust.xyz/r/opnform.md): Open source no-code form builder with logic and integrations
- [Firecrawl](https://wunderlust.xyz/r/firecrawl.md): API that turns web pages into Markdown or JSON for agents and apps
- [DocETL](https://wunderlust.xyz/r/docetl.md): Processes document collections with LLMs in steps that optimize themselves
- [Obsidian Local REST API](https://wunderlust.xyz/r/obsidian-local-rest-api.md): An Obsidian plugin that exposes the vault through a local API and MCP server
- [Open Notebook](https://wunderlust.xyz/r/open-notebook.md): An open source, self-hosted alternative to NotebookLM
- [pdfme](https://wunderlust.xyz/r/pdfme.md): Open source PDF generator with a visual template editor