--- title: "Klippa Integration" description: "Turn receipts and invoices into structured data with Klippa OCR. Agents extract totals, VAT lines, merchant names, and line items, classify documents by type before routing, and anonymize sensitive fields ahead of sharing or archiving." url: https://flowrunner.ai/integrations/klippa date_modified: 2026-08-01T02:40:32-07:00 --- # Klippa [Document Processing](https://flowrunner.ai/integrations/category/documents-forms) Turn receipts and invoices into structured data with Klippa OCR. Agents extract totals, VAT lines, merchant names, and line items, classify documents by type before routing, and anonymize sensitive fields ahead of sharing or archiving. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 3 actions · API key · available [Klippa website](https://custom-ocr.klippa.com/) · [Platform Documentation](https://custom-ocr.klippa.com) · Capability data verified 2026-07-27 1. An invoice PDF arrives as an email attachment in the accounts payable inbox 2. Classify Document confirms the file is an invoice rather than a statement, receipt, or marketing PDF 3. Parse Document extracts the total, currency, merchant, invoice number, purchase date, and VAT lines 4. The extracted merchant is matched against the vendor master, and the total is checked against the open purchase order 5. When the amount deviates from the PO, the vendor is new, or the extraction confidence is low, the AP lead reviews the original document beside the extracted fields before anything is booked 6. The approved payable is created in the accounting system with the source document attached 7. The finance channel logs the booking with merchant, amount, and match status ## What This Integration Enables Finance document intake is the canonical case for the FlowRunner orchestration model: the volume is relentless, the work is mechanical, and the cost of a wrong number is real money leaving the account. Klippa supplies the mechanical half. FlowRunner agents run OCR on receipts, invoices, and other financial documents supplied by URL or base64, extracting totals, currency, merchant names, invoice numbers, VAT lines, and individual line items. Classification identifies what a document is before a workflow assumes it, and anonymization masks names, addresses, bank account numbers, and ID numbers before a document is shared or archived. What Klippa does not supply is judgment about whether the extracted numbers should move money, and that is deliberate in how these workflows are built: extraction is autonomous, booking is gated through [human-in-the-loop](https://flowrunner.ai/concepts/human-in-the-loop) review where it counts. - Turn emailed and scanned invoices into structured payables without manual keying - Route documents by predicted type so invoices, receipts, and contracts follow different paths - Extract line items and VAT detail for expense policies and tax treatment - Mask sensitive fields before documents go to auditors, vendors, or long-term storage ### Without FlowRunner **Keying as a job function**: Someone retypes totals, dates, and VAT lines from PDFs into the ledger, one document at a time **Every document trusted equally**: The statement, the duplicate, and the real invoice all enter the queue looking identical **Sharing means exposure**: Sending documents to auditors or vendors risks leaking names, IBANs, and ID numbers ### With FlowRunner **Extraction at machine speed**: Totals, VAT lines, merchants, and line items become structured data on arrival **Documents routed by what they are**: Classification separates invoices from receipts and contracts before processing begins **Masking before movement**: Sensitive fields are anonymized before a document leaves the finance perimeter ## Use Case Scenarios ### The AP inbox that books itself, mostly Invoices arrive by email. For each attachment, the agent runs Classify Document, discards the statements and promotional PDFs, and runs Parse Document with the financial template on the real invoices. Clean extractions that match an open PO in [Xero](https://flowrunner.ai/integrations/xero) become draft bills automatically. Mismatched totals, unknown merchants, and low-confidence reads queue for the AP lead with the source document and extracted fields side by side. The team keys nothing and reviews only exceptions. ### Expense receipts from photo to ledger row Field employees photograph receipts on their phones. The agent runs Parse Document on each HEIC or JPG, extracting merchant, date, total, and VAT, and appends the structured row to the expense sheet in [Google Sheets](https://flowrunner.ai/integrations/google-sheets) with the image linked. Receipts that exceed policy thresholds or parse with missing fields go to the approver through [Slack](https://flowrunner.ai/integrations/slack). Month-end expense reconciliation starts from structured data instead of a shoebox. ### The audit package that leaks nothing An auditor requests a quarter's supplier documentation. Before anything leaves the finance perimeter, the agent runs Anonymize Document on each file to mask bank account numbers, personal names, and addresses that the audit scope does not require, then assembles the package. A reviewer spot-checks the masked set before release. The audit gets what it needs; the vendor's IBAN and the employee's home address do not travel with it. ## Human-in-Loop Highlight Parse Document's output is a payment instruction in waiting. The extracted total flows into a bill, the bill into a payment run, and a misread digit, a currency confusion, or a cleverly doctored invoice becomes money wired to the wrong place, which no workflow can wire back. So the booking step, not the extraction step, carries the gate: any invoice whose total deviates from its purchase order, whose merchant is absent from the vendor master, or whose extraction confidence falls below the team's threshold stops and waits for the AP lead, original document and parsed fields side by side. Anonymize Document gets the same respect at the perimeter, because a masked document that missed one IBAN is not masked once it has been sent. A person confirms the masking held before the package leaves. Extraction runs at machine speed all day; the moments that move money or expose data wait for a name in the audit trail. Agent processes routinely Detects exception requiring judgment Clear match Continues automatically Ambiguous Routes to human via preferred channel Human decides Agent resumes with decision ## Agent Capabilities 3 actions ### Document Parsing 3 - **Parse Document** Runs OCR on a receipt, invoice, or other financial document and returns structured data: total amount, currency, merchant name, purchase date, invoice number, VAT lines, and individual line items. Accepts a public URL or base64 content in GIF, PNG, JPG, HEIC/HEIF, or PDF, with a choice of parsing template to control returned fields. - **Classify Document** Classifies a document into its type, such as invoice, receipt, ID document, or contract, returning the predicted label with a confidence score. The routing step that keeps workflows from parsing a statement as an invoice. Requires a classification model enabled on the account. - **Anonymize Document** Masks sensitive information such as names, addresses, bank account numbers, and ID numbers, returning the anonymized result. The step that runs before any document leaves the finance perimeter. Requires an anonymization model enabled on the account. ## Frequently Asked Questions ### What can FlowRunner do with Klippa? FlowRunner agents can run Parse Document, Classify Document, and Anonymize Document in Klippa. ### Does connecting Klippa to FlowRunner require OAuth? No. Klippa connects to FlowRunner with an API key, no OAuth flow required. ### Can Klippa trigger a FlowRunner workflow automatically? Klippa doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at Klippa?** This integration exposes Klippa to AI agents on every FlowRunner plan, including through MCP, at no cost to you. [See what FlowRunner offers integration partners](https://flowrunner.ai/integrations/partners), including how to keep this page current. --- Markdown version of https://flowrunner.ai/integrations/klippa. Site index: https://flowrunner.ai/llms.txt