Write and send Gmail from a plain request

What it is

This n8n workflow is a tool for other workflows to call, and it sends real email. Whatever calls it passes one input named query, a request in plain words. OpenAI (gpt-4o) turns it into a recipient, a subject and a message that ends with a sign-off, the Gmail node sends it as plain text, and the last node sets response to Sent. Four nodes in all.

Who it suits

Builders of n8n AI agents who want a send-an-email tool, and teams that send routine messages and would rather say it in a sentence. Skip it if a person should read each email before it goes. Nothing here asks for approval.

Where to use it

Inside a larger workflow, as the target of an Execute Workflow node or as a tool an AI agent can call. Any n8n works, as long as a Gmail account can be authorized in it.

Setup

Get it saves a zip with workflow.json and Mindket’s setup notes in your downloads folder. Its author puts setup at about 9 minutes.

1. In n8n, create a workflow, open its menu, choose Import from File and pick workflow.json.
2. Add an OpenAI credential with your key and select it in OpenAI.
3. Add a Gmail credential, sign in with Google, and select it in Gmail.
4. Replace YOUR_NAME in two places: the Sender Name option of Gmail, and the sign-off line in the system message of OpenAI.
5. Save and publish it, then open the calling workflow and choose this one in its Execute Workflow node or tool.

The trigger declares one field, query, of type any. The caller passes this and gets one field back:

{
  "query": "Email dana@example.com a reminder that the invoice is due on Friday"
}

{
  "response": "Sent"
}

How to use it

Write the request as you would tell an assistant: who it goes to, what it is about and the tone. OpenAI returns eight fields, and Gmail uses three of them: sendTo, subject and message. The other five (emailType, appendAttribution, ccList, bccList, senderName) are asked for but never reach Gmail, which has its own fixed settings: plain text, no n8n attribution, and the sender name in the node. A CC or BCC in the request is parsed and then dropped, unless you wire those two lists into the Gmail node’s options.

Example

A caller passes the request from the fence above. The model sets sendTo to the address, writes a short subject about the invoice and a polite message that ends with the sign-off, and Gmail sends it from your account. The caller receives Sent.

For newcomers and experts

If you are new. A tool workflow runs only when another workflow calls it, and this one sends for real. Test it with your own address as the recipient before anyone else’s.

For experts. The OpenAI node runs gpt-4o with JSON output on. Its user message is JSON.stringify($json.query), so a string or an object both arrive as text. The system message names the eight fields and tells the model to sign off with a first name and never to leave a placeholder such as a bracketed name. Gmail reads the model’s answer from message.content, and Sent comes back only once Gmail has accepted the message.

Good to know

The risk is in what the model writes, because nothing checks it before it goes out.

• No review: a wrong recipient or an invented detail in the body is sent as written, so for an agent add a confirmation step before this tool
• Sender: the mail leaves your authorized Gmail account, with the name you set
• Billing: one OpenAI request per email, and Gmail’s own sending limits apply
• Privacy: the whole query goes to OpenAI
• Credentials: the JSON holds none, and the author’s own name is replaced by YOUR_NAME