Parseur MCP. Turn Invoices, Receipts, and Emails Into JSON.

You're building an AI agent that needs clean data from messy documents—invoices, receipts, emails. You don't want your agent slogging through raw PDFs or unstructured text. This server lets you connect document extraction pipelines straight into your workflow. It handles the gnarly OCR and templating stuff; all you gotta do is feed it files and get structured JSON back.

Setting Up Your Document Intake Pipeline

First, you need a place for the documents to go. You can start by listing all active pipelines using list_mailboxes so you know what's running. To create a new intake point—say, one just for invoices or another just for web forms—you use create_mailbox. This tells the system exactly what kind of files it should handle: PDFs, emails, or attachments.

You can check out the full setup details later using get_mailbox if you gotta verify how things are configured.

Next up is telling the server what to look for. Defining rules means creating a template that maps specific fields from any incoming document. When you use create_template, you're defining precisely what data points you expect—like the total amount, an invoice number, or a date range. You can list all existing templates tied to a mailbox using list_templates to make sure your rules are set up right.

Feeding the Documents and Running the Job

Once the pipeline is ready and the template is locked down, you feed it the goods. Sending a file URL directly into a specified mailbox starts the whole automated parsing process. You use upload_document for this; the document hits a queue, and the OCR engine gets to work.

You gotta keep an eye on what's happening. To see every document that's been added, you run list_documents. This gives you the ID, basic metadata, and most importantly, the status: Is it pending? Did it fail? Or is it fully processed?

If a job fails or throws an error, don't rewrite the whole thing. You use retry_document to force Parseur to re-run the parsing on that existing document ID. This works best after you’ve tweaked the underlying template rules.

Retrieving and Using the Data

When a document status is 'processed,' you can pull out the clean data. You use get_document_data to retrieve a final, structured JSON dictionary containing every mapped field—you'll get things like { "invoice_number": "A123", "total_amount": 500.00 }. This is the raw material you plug right into your database or workflow engine.

If you just need to know when the document arrived or who sent it, but don't care about the actual extracted data yet, get_document_details grabs that metadata for you. It keeps the process clean and targeted.

This server handles the whole loop: from creating a dedicated intake point (create_mailbox) to defining strict rules (create_template), pushing files in (upload_document), monitoring job status (list_documents), fixing errors (retry_document), and finally, spitting out perfectly structured JSON data (get_document_data). You get clean properties without ever having to manually copy-paste anything.