Everything here is about the plumbing — point a source at a destination, run the workflow, then check on the job — which is what you want when the ingestion into a vector database has to be set up or re-run rather than debugged one file at a time. Note the repository's own header: it is kept for reference, and Unstructured directs parsing work to its hosted Transform server instead.
A control-plane server for the Unstructured API: it manages source connectors, destination connectors, workflows and the jobs they produce. The repository states it is no longer actively maintained and points to the hosted Unstructured Transform server for parsing files.
- Source connectors listed, inspected, created, updated and deleted
- Destination connectors handled the same way, covering vector databases and storage targets
- Workflows created against a source and destination, updated, run on demand and deleted
- Jobs listed per workflow, inspected individually and cancelled
- A listing of workflows that have completed at least one job, with the source and destination details attached
An Unstructured API key from platform.unstructured.io, set as UNSTRUCTURED_API_KEY, and Python 3.12+ with uv. Published on PyPI as uns_mcp and run with uvx over stdio. Each connector you create needs that service's own credentials in the environment — AWS_KEY and AWS_SECRET for S3, PINECONE_API_KEY for Pinecone, ASTRA_DB_APPLICATION_TOKEN with its endpoint for Astra, and so on for Azure, Databricks, OneDrive, SharePoint, Neo4j, MongoDB, Weaviate, Google Drive and Salesforce.
One command plus a key — uv pip install uns_mcp, then supply credentials