Gmail invoice PDFs read by Cradl AI into Google Sheets

Gmail invoice PDFs read by Cradl AI into Google Sheets

An n8n workflow that watches Gmail for invoice emails and sends each PDF attachment to a Cradl AI agent. It waits for the agent's answer, which can include a person's review of the values the agent was unsure about. Then it adds one Google Sheets row for each invoice line.

What is inside

• workflow.json: the workflow as n8n exports it, 10 nodes. 6 of them run and four are sticky notes. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• The agent in Extract invoice details with AI was the original author's own. It is now <YOUR_CRADL_AGENT_ID>, and you choose your agent from the list.
• The two settings that node shows by default, the operation and Wait for Results, are now written into the file.
• Add invoice line to Google Sheets had no spreadsheet. It now holds <YOUR_SPREADSHEET_ID>.
• Split out each invoice line puts each line under a field called line. Before, it used a field with dots in its name, which is awkward to map.
• The Code node Filter PDF attachments used a loop variable without declaring it and assumed every email had attachments. It also did not tell n8n which email each PDF came from, and Mark a message as read needs that. It now does all three correctly.
• The sticky notes no longer link to the vendor's sign-up and video pages, and they say what each step needs.
• Nothing else. The Gmail filter and the poll every minute are as they were.

Before you start

• n8n with the Cradl AI community node installed. It is not a node that n8n ships, and its latest version on the npm registry on 12 November 2026 was 1.0.9. Your n8n must also be reachable from the internet, because the node hands Cradl AI a resume address for each run.
• A Cradl AI account and an agent that reads invoices, with at least one test document. Cradl AI's pages describe the agent, the review screen and the n8n setup.
• A Gmail account and a Google account, each connected to n8n with its own credential.
• A Google spreadsheet for the rows. An empty sheet is fine.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. In Cradl AI, open your agent, add a trigger and choose n8n. Copy the Client ID and Client Secret from that panel.
3. Open Extract invoice details with AI and click Set up credential. Paste the two values, then choose your agent under Agent Name or ID.
4. Open Gmail Trigger, connect your Gmail account and check the search filter. It reads has:attachment subject:invoice for unread email. For a shared mailbox you could add to:<YOUR_INVOICE_ADDRESS>.
5. Open Add invoice line to Google Sheets. Connect your Google account, choose your spreadsheet and sheet, and map each column. A value comes from {{ $json.line.<field id>.value }}.
6. Run the Gmail trigger once with a test email, then open the output of the Cradl AI node. The line items sit under body.output.<group id>. If your group is not called invoice_lines, put its id in the field of Split out each invoice line.
7. Save and press Publish at the top of the canvas (or Shift+P).

How a run goes

1. Gmail Trigger checks every minute for unread email that matches the filter, and downloads the attachments.
2. Filter PDF attachments makes one item for each PDF, with its file name.
3. Mark a message as read marks the email read. This happens at once, in parallel with the next step.
4. Extract invoice details with AI uploads the PDF to your agent and pauses the run. When the agent is done, and any review has finished, Cradl AI calls the run's resume address. The node then outputs body, with output and context, and the document as binary data.
5. Split out each invoice line makes one item for each entry in body.output.invoice_lines, keeps the other fields and puts the line under line.
6. Add invoice line to Google Sheets appends each line as a row.

Change the look

• Which emails are read: the search filter in Gmail Trigger.
• Which files are kept: Filter PDF attachments. Change "pdf" to keep another type.
• Which fields are read and which values go to review: the agent in your Cradl AI account.
• Where the rows go: Add invoice line to Google Sheets. You can swap it for a node that sends them to an accounting tool or a database.

Good to know

• Not run: Mindket made no call to Cradl AI, Gmail or Google. It read Cradl AI's pages, the Cradl AI node's code at version 1.0.9 and n8n's own node code on 12 November 2026, and it ran Filter PDF attachments on sample items on its own machine.
• The Google Sheets node has no column mapping, so it stops with an error until you map the columns in the n8n editor.
• Cradl AI's webhook page shows each field as an object with value, rawValue, confidence, page and name, grouped under ids. Use .value for the normalised text and .rawValue for the original.
• The email is marked read before Cradl AI has answered. If a run fails, that email is not tried again, so check failed runs in Executions.
• A run that waits for a result stays in Waiting until Cradl AI calls back, so a long review leaves the run open for that time.
• Cradl AI's page says the agent learns from each human correction to improve the model. The workflow does not do this, because it happens inside Cradl AI.
• Pricing and plan limits for Cradl AI were not checked, so look at your plan before you run it on a mailbox.
• Your data: each PDF goes to Cradl AI's servers, and the rows go to Google. Invoices can hold names, addresses and bank details.
• Credentials: the JSON holds none. Gmail, Google Sheets and Cradl AI each need their own n8n credential.