MailMint for n8n

Publication status #

Install #

Today — build the tarball and install that

Until the package is published this is the honest route, and it is the one every screenshot on this page was taken through:

npm install --ignore-scripts
npm run build
npm pack                       # -> n8n-nodes-mailmint-<version>.tgz

cd ~/.n8n/nodes
npm install /path/to/n8n-nodes-mailmint-<version>.tgz

Then restart n8n.

Once it is published

Settings → Community Nodes → Install, then enter n8n-nodes-mailmint; or cd ~/.n8n/nodes && npm install n8n-nodes-mailmint from the CLI. Neither works yet — see above.

Once installed, both nodes appear in the node panel under their own names. The trigger is found by searching the trigger panel for MailMint:

The credential #

1. In n8n, add a MailMint API credential.
2. API Key — the key for your MailMint account. It starts with mm_live_.
3. Base URL — the root URL of the MailMint API you are talking to, with no trailing slash. There is no default, because there is no hosted MailMint to default to. Use your own deployment, or http://host.docker.internal:3100 when the API runs locally and n8n is in Docker.
4. Save.

The credential tests itself against GET /v1/usage, so you get a real green tick before you build anything rather than an “untested” badge:

First workflow — mail you already have #

You do not need a MailMint address to start, which matters, because the hosted address is not live yet. If mail is already reaching n8n — through the built-in Email Trigger (IMAP), through Gmail, through anything — put a MailMint node after it and it parses what arrives.

1. Add an Email Trigger (IMAP) node and point it at your mailbox. Any Format works.
2. Add a MailMint node after it. It opens on Parse → Parse Email with Input: Automatic, which is already correct.
3. Under Fields, click Add Field three times and fill them in.
4. Execute step.

The schema editor #

This is the part that does not exist anywhere else. Each field you add has five inputs:

A value that cannot be coerced comes back as null with a type_error flag. It is never invented, and never a guess string like “N/A”.

Schema also takes From a Mailbox, to reuse the schema saved on one of your MailMint mailboxes, and JSON, to compute the whole schema in an earlier node.

What comes out #

{
  "invoice_number": "INV-2291",
  "total": 31.5,
  "due_date": "2026-09-08",
  "_needs_review": false,
  "_meta": {
    "subject": "Invoice INV-2291 from Acme Ltd",
    "from_email": "billing@acme.example",
    "received_at": "2026-08-25T09:14:03.221Z",
    "type": "invoice",
    "needs_review": false,
    "flags": [],
    "attachment_count": 1,
    "attachment_names": ["invoice-2291.pdf"]
  }
}

That is Simplify, which is on by default, because the next node is nearly always a Google Sheets, a Postgres or an IF. Turn it off and you get the full message object: headers, both bodies, tables, detected amounts and dates, and per-field confidence with the evidence each value was read from.

Line items — one item per row #

The loudest complaint in this whole category is “the mail had forty rows and I got one”. Set Output to One Item Per Line Item and one email becomes one n8n item per invoice line, with the header fields repeated on each.

{ "invoice_number": "INV-2292", "total": 132,
  "Item": "Widget", "Qty": "3", "Amount": "$27.00",
  "_row_index": 0, "_row_count": 3, "_line_items_truncated": false }

• It finds the rows on its own: an array field you defined first, then the largest table found in the body. Name a specific one in Line Items From if you want a particular table.
• _row_count travels on every row and _line_items_truncated says so out loud when the table came back short. A short array can never be silent.
• A message with no rows still produces exactly one item, with _row_count: 0. Nothing is ever silently dropped.
• pairedItem is set on every row, so n8n can still trace all forty items back to the one email they came from.

Routing what needs a human #

Every field carries a confidence, the source it came from and the verbatim evidence it was read out of. Turn on Options → Include Confidence:

"_confidence": {
  "total": { "confidence": 0.97, "source": "rule+llm", "evidence": "Total: $31.50" }
}

_needs_review sits at the top level of every simplified item. Turn on Route Messages Needing Review Separately and the node grows a second output: anything with a missing required field, a low-confidence value, a type it could not coerce, or evidence it could not find in the mail goes down Needs Review instead of Parsed.

Line items follow their message: forty rows from a doubtful invoice all go down the same branch together, rather than being split across two.

The same setting exists on the trigger, so mail that needs a human can be routed the moment it arrives.

Attachments #

Filenames, content types, sizes and checksums are on every simplified item already, under _attachments, with no option to turn on. The bytes are the one thing behind a switch (Options → Include Attachment Bytes), because a single PDF is usually larger than the rest of the message put together. To get a file as real n8n binary data, use Message → Download Attachment.

MailMint Trigger #

For mail that should come to MailMint rather than through your own IMAP. One node, two modes.

Webhook — the default

It registers itself. On activation it sets the mailbox’s webhook_url to this workflow’s URL and installs a signing secret it generates for you; on deactivation it clears them. Every incoming delivery has its x-mailmint-signature HMAC verified before the workflow runs — a body that does not verify is answered 401 and never starts anything. Instant, and there is nothing to configure beyond picking the mailbox.

Options → Signature Tolerance (Seconds) defaults to 300, and Options → Webhook Secret lets you pin your own instead of letting the node generate one.

Polling

For an n8n the internet cannot reach. It reads GET /v1/events on your poll schedule and remembers the cursor on the node. On first activation it seeds the cursor and emits nothing, so switching a workflow on never floods it with a week of old mail.

Filters

Both modes take the same Filters — Needs Review Only, From Sender (an address or a domain) and Mailbox Name or ID — and the same Output, Simplify and Route Messages Needing Review Separately settings as the main node.

All eleven actions #

The main node is marked usableAsTool, so an n8n AI Agent can call MailMint directly as a tool.

Errors #

Every failure carries the API’s own message and its hint, plus the item index that failed. With Settings → Continue On Fail on, the failing item becomes:

{ "error": { "code": "message_not_found", "message": "…", "hint": "…", "httpCode": "404" },
  "errorMessage": "…" }

so an IF node can branch on {{ $json.error.code }} instead of matching on a sentence. The full code list is in the API reference.

What works on n8n Cloud today #

n8n Cloud only accepts verified community nodes, and n8n-nodes-mailmint is not published at all yet, let alone verified. Until it is, an HTTP Request node does the same job from any n8n:

Method:            POST
URL:               {{ $env.MAILMINT_URL }}/v1/parse
Authentication:    Generic → Header Auth
  Name:            Authorization
  Value:           Bearer mm_live_…
Send Body:         on   (JSON)
Body:
  {
    "subject": "{{ $json.subject }}",
    "text":    "{{ $json.textPlain }}",
    "html":    "{{ $json.textHtml }}",
    "schema": [
      { "name": "invoice_number", "type": "string", "description": "the invoice or reference number" },
      { "name": "total",          "type": "number", "description": "grand total including tax" }
    ]
  }

You lose the schema editor, the line-item fan-out and the second output — everything else, including the confidence and evidence on every field, is identical, because the node is a thin client over the same endpoint.