Unstructured MCP. Automate Data Ingestion from Any Source to Vector DB.

Listen up. This server handles your entire raw data lifecycle, taking messy documents—PDFs, reports, whatever—and turning them into clean, structured data your AI client can actually use. You connect this thing to your agent so it can automate document ingestion pipelines without you having to open some clunky dashboard and mess around with settings.

It’ll pull docs from sources like AWS S3 or Google Cloud Storage, let you define the processing rules, and spit out clean records straight into Vector DBs or SQL tables. Your AI agent becomes a command center for building and running Retrieval-Augmented Generation (RAG) pipelines using real data.

Here's what it does:

Listing Data Sources and Targets: You can check where your documents sit and where the clean output needs to go. The list_data_sources tool shows you every configured remote connector, whether that’s an AWS S3 bucket or a Google Cloud Storage location. Similarly, if you need to know what kind of databases are waiting for data, the list_data_destinations tool displays all target locations—that means specific Vector DB endpoints and SQL table definitions.

Managing Pipelines: To see what's possible, you first gotta list out every defined end-to-end pipeline. The list_processing_workflows function gives you a rundown of every existing document processing workflow. Once you know which pipelines exist, the get_workflow_details tool lets you pull up the precise configuration details for any specific one; it shows exactly how that data transformation is supposed to happen.

Running and Tracking Jobs: You don't wanna wait around watching things happen. The server lets you manually kick off a whole workflow run immediately using trigger_workflow_execution, and it returns a job ID so you know what started. To keep tabs on that, the list_workflow_jobs tool shows you both a history of completed tasks and any jobs that are still running.

This log tells you if the sync finished successfully or if there was an error in the processing task.

This whole setup means you can manage data ingestion from source discovery—using list_data_sources to find your files on S3 or GCS—all the way through defining the rules with get_workflow_details, triggering the job with trigger_workflow_execution, and finally confirming the clean output landed exactly where it needed to go, listing those destinations via list_data_destinations.

It's a full loop: find it, define how to process it, run it, check its status.