Use Inbox Triage with n8n

Add the node

1. Add a Schedule Trigger node, or start the workflow manually while testing.
2. Add an MCP Client node for a plain, fixed tool call with no AI model involved.
3. MCP Endpoint URL: https://getinboxtriage.com/mcp.
4. Server Transport: HTTP Streamable.
5. Choose your authentication (see below).
6. Tool: pick show_inbox from the list n8n fetches from the server. Input Mode: JSON, with {"filter": "needs"} as the input.

For a workflow where an AI model decides which tool to call, use an AI Agent node instead, and add an MCP Client Tool sub-node under its Tools with the same endpoint, transport, and authentication. Set Tools to Include to All Except, and exclude send_draft, confirm_mark_spam, confirm_unsubscribe, set_send_enabled, and start_imap_secret_capture.

Sign in with OAuth

Choose Authentication: MCP OAuth2, create a new MCP credential, and leave Use dynamic client registration turned on with the Authorize and Token URL fields blank: n8n discovers them from the server on its own (dynamic client registration needs n8n 1.119.0 or later). Sign in to Inbox Triage and choose Allow. n8n's callback path is already accepted on our side, and n8n refreshes the token on its own after it expires.

Or use an API key

Create a key in Settings, under Advanced (see API keys). Choose Authentication: Bearer Auth, create a new credential, and paste the key as the token.

Your first workflow

Import the ready-made template: n8n-inbox-triage.json. In n8n, open Workflows, then Import from URL, and paste that address (or download the file and use Import from File). It has a Schedule Trigger, an MCP Client node calling show_inbox with filter: needs, and a placeholder Set node for the result. Open the MCP Client node and add your own Bearer Auth credential; if the Tool field shows an error, reselect show_inbox from the Tool list, since n8n needs a live connection to fetch it. Replace the Set node with a Slack node, or anything else, to post the result's sayToHuman line somewhere you will see it.

We shipped this as a plain MCP Client node rather than an AI Agent with an MCP Client Tool, since a fixed daily check does not need a language model deciding which tool to call, and it needs no extra AI provider credential.

What it will not do on its own

n8n has no way to show Inbox Triage's in-chat approval prompt, so a connection through n8n is always treated as unattended, whether it signs in with OAuth2 or Bearer Auth. Calling send_draft, confirm_mark_spam, or confirm_unsubscribe from n8n never sends, reports spam, or unsubscribes on its own. The tool's result carries an approveUrl link instead: only the signed-in person's click on that page, on getinboxtriage.com, completes the action.

Known issues

• Older n8n versions had a bug where the transport dropdown showed HTTP Streamable while the node silently kept using SSE (n8n issue #18938), reportedly fixed in later versions. Confirm your n8n version behaves correctly before relying on it.
• An open community report describes the MCP Client Tool timing out on HTTP Streamable while working on SSE, with no confirmed fix date. If calls to Inbox Triage time out, try the older SSE transport as a stopgap and let us know at help@getinboxtriage.com.
• Some n8n agent versions open a new HTTP Streamable session on every single tool call rather than reusing one, so each call to Inbox Triage can look like a brand new client. This does not break anything on our side, but it means our server never sees a long-lived n8n session.
• Past versions of n8n's dynamic client registration sent the wrong redirect URI back to the server. If OAuth2 sign-in fails, check that the redirect address on your credential screen matches what n8n itself is showing you.
• One reported bug caused a workflow that ignored the transport dropdown to retry a failing call tens of millions of times against one server. Watch your request volume the first time you turn on a scheduled n8n workflow against Inbox Triage.

Can I connect two Gmail accounts through n8n?

Yes. Connect both inboxes in Inbox Triage first. One n8n MCP Client or MCP Client Tool node can reach both, since the inbox is picked inside each tool call, not by the connection itself.

Does n8n need my Gmail password?

No. n8n only ever gets a bearer key or an OAuth2 token for Inbox Triage. Your mail passwords and provider sign-in tokens stay with Inbox Triage.

Which tools should I exclude from an AI Agent?

Exclude send_draft, confirm_mark_spam, confirm_unsubscribe, set_send_enabled, and start_imap_secret_capture from the MCP Client Tool sub-node so an approval link, not the agent, decides when those happen.

HTTP Streamable keeps timing out for me. What do I do?

Check your n8n version first, since older versions had real transport bugs. If it still happens, SSE is a working stopgap even though the MCP standard now prefers Streamable HTTP.

Do I need the AI Agent node at all?

No. The plain MCP Client node calls one named tool with fixed arguments and needs no language model, which is enough for a scheduled check like the template.