Invoice OCR to Postgres (or MySQL) with n8n

What we're building

Spreadsheets are fine for pilots. Finance systems want typed rows in Postgres or MySQL. n8n ships native Postgres and MySQL nodes — you extract with Cryvis, then INSERT without bouncing through Airtable as a middle database.

Google Drive Trigger
        |
        v
Download → $binary.data
        |
        v
HTTP Request
  POST https://api.cryvis.com/v1/documents/invoice
  Form-Data: file = data
        |
        v
IF {{ $json.success }} equals true
        |
        v
Postgres (or MySQL): Insert
  invoice_number, dates, seller, totals, source_file

Compared to a Zapier → Airtable path: one less SaaS copy of vendor tax IDs, SQL constraints, and joins to your ERP staging tables. Sheets version of the same extract: Extract invoice data. Hub: /blog/n8n.

What you need

• n8n with Postgres or MySQL credentials (Cloud or self-hosted; self-hosted often sits on the same VPC as the DB)
• Drive folder for inbound invoices
• Cryvis API key — Invoice API · extractInvoice
• Table DDL below (or adapt)

Credits: 1 per page/image.

Suggested schema (Postgres)

CREATE TABLE invoices (
  id              BIGSERIAL PRIMARY KEY,
  invoice_number  TEXT,
  invoice_date    DATE,
  due_date        DATE,
  currency        TEXT,
  seller_name     TEXT,
  seller_tax_id   TEXT,
  buyer_name      TEXT,
  subtotal        NUMERIC(14, 2),
  tax_total       NUMERIC(14, 2),
  total_amount    NUMERIC(14, 2),
  amount_due      NUMERIC(14, 2),
  source_file     TEXT,
  drive_file_id   TEXT,
  extracted_at    TIMESTAMPTZ DEFAULT NOW(),
  UNIQUE (seller_name, invoice_number)
);

MySQL: same columns; use AUTO_INCREMENT and DATETIME(6) as you prefer. The unique key stops vendor resends from double-inserting.

Workflow nodes

1. Google Drive Trigger (or Schedule + list new files)
2. Google Drive — Download → $binary.data
3. IF — MIME / extension allowlist (optional but recommended)
4. HTTP Request — Cryvis
5. IF — {{ $json.success }} equals true
6. Postgres / MySQL — Insert (or Execute Query with parameterized INSERT)
7. Optional: Google Drive — Move to Processed; false branch → error log table

Cryvis HTTP Request

Full multipart notes: HTTP Request OCR setup.

Map expressions into INSERT

Use the Postgres node’s column mapping UI, or Execute Query:

INSERT INTO invoices (
  invoice_number, invoice_date, due_date, currency,
  seller_name, seller_tax_id, buyer_name,
  subtotal, tax_total, total_amount, amount_due,
  source_file, drive_file_id
) VALUES (
  $1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13
)
ON CONFLICT (seller_name, invoice_number) DO NOTHING;

Wire query parameters from the same {{ $json.data.* }} expressions. Prefer ON CONFLICT DO NOTHING (or MySQL INSERT IGNORE / ON DUPLICATE KEY) over silent duplicate finance rows.

Null and type pitfalls

• Dates may be null if OCR cannot read them — allow NULL columns or route to an invoices_review table via IF {{ $json.data.invoice_number }} is empty.
• Numbers from Cryvis are JSON numbers — map to NUMERIC, not text with currency symbols.
• seller.tax_id is often null; that is valid.

Error handling

HTTP Request
  retry on fail: 2–3 (5xx only in practice)
        |
        +-- IF success false → INSERT into invoice_errors (file, status, body snippet)
        |
Error Trigger workflow → Slack: workflow + node + message (no full PII dump)

Self-hosted: put the DB credential on a restricted network path; do not expose Postgres to the public internet just for n8n Cloud if policy forbids it — run n8n in the same private network instead.

Test

1. Insert one fixture PDF into Drive.
2. Run workflow; confirm one row in invoices.
3. Re-drop the same file (or re-run) — unique constraint / ON CONFLICT should block a second row.
4. Upload a PNG logo — MIME IF should stop before HTTP (no wasted credits).
5. Spot-check amount_due and currency against the PDF.

Why not Sheets here

Sheets is fine for ops review. Databases win when you need:

• Unique constraints on (seller, invoice_number)
• Joins to purchase orders / vendors
• Downstream dbt or ERP loaders reading SQL
• Retention and access control that match your security review

Keep a parallel Sheets append only if humans need a live queue — or query Postgres from a BI tool.

Production checklist

• Unique key on seller + invoice number
• Move Drive file after successful INSERT
• Error table + Error Trigger alert
• Credit budget for multi-page statements — Credits
• Credential rotation for Cryvis Bearer and DB user

Next steps

Gmail → Sheets + Slack triage: Automate invoice processing. Building-block HTTP: HTTP Request OCR.

Sign up · Pricing · Invoice · extractInvoice · /blog/n8n.