Unstructured
An open-source library and MCP server that turns PDFs, Word documents, HTML, images and other unstructured files into clean, structured data ready for language models, vector databases and other downstream processing.
Unstructured provides open-source components for ingesting and pre-processing unstructured documents, such as PDFs, HTML and Word files, into structured data for language model applications. Its modular functions and connectors handle ingestion and pre-processing across different source platforms.
Unstructured Transform exposes the same document processing as an MCP server for AI agents, working with clients such as Claude Code, Cursor and Codex CLI. An agent authenticates once, is pointed at a local file or URL among more than 60 supported formats, and can be told in plain language what to do with it, such as parsing and chunking a document for a vector store, with the server handling partitioning, enrichment, chunking and embedding.
A separate hosted product, Unstructured Pipelines, moves this processing to production with additional chunking, embedding and enrichment features through a low-code UI or an API.