DocuPipe

@DocuPipe find all invoices uploaded in the last 7 days and show me their processing status

Found 14 documents in the 'invoices' dataset from the past week. 12 are fully processed (status: completed), 1 is still extracting tables (status: processing, 87% complete), and 1 failed due to an unsupported file format (status: error). The completed documents include invoice_2024_Q1_final.pdf (42 pages, extracted 3 days ago) and vendor_statement_march.pdf (8 pages, extracted yesterday). The failed document was a .pages file that requires conversion to PDF before upload.

This example shows DocuPipe's document-browsing capability filtered by dataset and time range. Users should know that processing status reflects the extraction pipeline's current state—documents may move from 'processing' to 'completed' between queries. The MCP requires read access to your DocuPipe account and will only surface documents your API key can view.

@DocuPipe create an extraction schema for purchase orders that captures vendor name, PO number, line items with quantities and prices, and total amount

Created extraction schema 'purchase_order_v1' (ID: schema_8x3k9m). The schema defines four top-level fields: vendor_name (string), po_number (string), line_items (array of objects with item_description, quantity, unit_price), and total_amount (number). You can now reference this schema when submitting documents to automatically extract structured data. The schema uses JSON Schema draft-07 and includes validation rules—quantities must be positive integers, prices must be non-negative numbers.

Schema creation is a one-time setup that unlocks structured extraction across all matching documents. This example demonstrates DocuPipe's ability to define reusable templates for domain-specific documents. Be aware that schema creation consumes credits, and poorly designed schemas may produce low-confidence extractions—test with sample documents before processing large batches.

@DocuPipe get the extracted data from document doc_7h2p5n and summarize any discrepancies between the stated total and the sum of line items

Retrieved document doc_7h2p5n (vendor_invoice_march_2024.pdf, 3 pages). Extracted data shows 8 line items totaling $4,287.50, but the invoice footer lists a total of $4,387.50—a $100 discrepancy. Line item 6 ('Consulting Services - March') is listed at $850.00, but the per-hour rate ($175/hr × 5 hrs) should yield $875.00, accounting for the $25 under-calculation. The remaining $75 gap is unexplained and may indicate a manual adjustment, tax line omission, or OCR error in the footer region.

This synthesis example pairs DocuPipe's structured extraction with the AI's analytical reasoning to surface data-quality issues. The MCP returns raw extracted fields; the AI performs the arithmetic and logic. Users should verify discrepancies against the source PDF, as OCR confidence varies by document quality—extraction confidence scores are available in the full document metadata.

When DocuPipe wins for customer onboarding workflows

A 6-person customer success team onboards 20-30 enterprise clients per quarter, each submitting contracts, compliance forms, and account setup PDFs. DocuPipe's schema-driven extraction turns those documents into structured JSON that feeds directly into Salesforce and the internal CRM. The team creates one schema per document type (W-9, MSA, security questionnaire), then runs batch jobs against incoming uploads. The Find Jobs and Get Account Usage tools let the CS lead track credit burn and spot bottlenecks before they block onboarding SLAs. This setup breaks down if your document types change weekly or require human-in-the-loop review on every extraction—DocuPipe assumes stable schemas and automated pipelines. If you're processing fewer than 50 documents a month, the API overhead probably isn't worth it.

Why DocuPipe fits AP automation at mid-market scale

A 3-person finance team at a 40-employee SaaS company receives 80-120 vendor invoices monthly via email and Slack. They built a Switchy workflow that watches a shared inbox, submits PDFs to DocuPipe with a single invoice schema, and writes extracted line items (vendor, amount, due date, GL code) into a Google Sheet for approval. The Delete Document tool cleans up processed files to stay under the plan's storage cap. The schema includes extraction guidelines that handle common vendor quirks (net-30 vs. due-on-receipt language, tax line variations). This works because invoice formats are predictable and the team can tune one schema over time. If your vendors send wildly inconsistent formats or you need real-time extraction (sub-10-second turnaround), DocuPipe's asynchronous job model will frustrate you.

When DocuPipe doesn't fit real-time support flows

A 5-person support team wants to auto-extract error logs and screenshots from customer tickets to speed up triage. DocuPipe's job-based architecture introduces 15-60 second latency between submission and parsed output, which kills the live-chat experience where agents need answers in under 10 seconds. The Find Jobs polling loop adds complexity to a Switchy workflow that should feel instant. DocuPipe shines when you can batch-process attachments overnight or during off-peak hours, then surface insights in morning standup—not when a customer is waiting on Intercom. If your support volume is under 200 tickets a week and attachments are mostly screenshots (not dense PDFs), a simpler OCR tool with synchronous responses will serve you better. Save DocuPipe for back-office document pipelines where speed matters less than extraction accuracy.