mdkit vs Unstructured
Unstructured is an ETL platform for LLM data with connectors and enterprise features; mdkit is a small document-to-Markdown API. Compared honestly.
Last updated
| Unstructured | An ETL platform for LLM data: partitioning, chunking, embedding and connectors to dozens of sources/destinations |
|---|---|
| mdkit | A single-purpose conversion API: documents in, Markdown/JSON out — bring your own pipeline |
Unstructured is an ETL-for-LLM platform: it partitions many file types into structured elements and, on its hosted platform, adds source and destination connectors (S3, SharePoint, vector databases, …), chunking, embedding and enterprise features. An open-source library covers the core partitioning; the platform is a commercial product aimed at organizations.
mdkit covers one step of that pipeline — turning a document into clean Markdown or structured JSON over a REST/MCP API — and deliberately nothing else. Your chunking, embedding and storage stay yours.
When Unstructured fits
You want a managed end-to-end ingestion pipeline — connectors on both sides, scheduling, enterprise controls — and a vendor relationship sized for that scope.
When mdkit fits
You already have (or want to own) the pipeline and just need the conversion step to be reliable: one small API with a free tier, flat subscription pricing, async jobs with signed webhooks, and an MCP server so agents can call it directly. Small enough to adopt in an afternoon and to drop just as easily. Try it free.
FAQ
- Is mdkit an Unstructured alternative?
- For the conversion step only. Unstructured is an ETL platform — partitioning, chunking, embedding, and connectors to dozens of sources and destinations. mdkit converts documents and stops there.
- Does mdkit chunk documents or generate embeddings?
- No, deliberately. It returns Markdown or JSON and leaves chunking and embedding to your pipeline, where you can change strategy without changing vendor.
- When is Unstructured the better fit?
- When you want the whole ingestion pipeline — connectors, chunking and embedding — as one managed product rather than assembling it yourself.