Outbound Email for AI Agents in Intercom

Intercom handles conversational support. MultiMail adds the approval gating, identity verification, and audit trail your AI agents need before sending email outside the thread.

Intercom's Fin AI agent and custom bots are effective at resolving support conversations — but when an agent needs to send a follow-up email outside the Intercom thread, you lose the oversight controls your team depends on. The message leaves through a shared mailbox with no approval step, no cryptographic proof of origin, and no audit trail.

MultiMail fills that gap. By routing agent-authored outbound email through MultiMail's REST API, you get per-message approval gating, DKIM-signed delivery from a verified sender identity, and a complete decision log — all without changing how Intercom handles inbound conversations.

The integration is a webhook-to-API bridge: Intercom fires a webhook when an agent drafts an outbound message, your backend calls MultiMail's `POST /v1/mailboxes/{mailbox_id}/send` or `POST /v1/oversight/decide` endpoints, and the approval state flows back to Intercom via its Conversations API. Your human team approves in MultiMail's queue; the message sends only after explicit sign-off.

This pattern applies to any Intercom workflow that touches outbound email: churn-risk escalations drafted by Fin, onboarding sequences triggered by lifecycle events, or proactive outreach composed by a custom agent reading CRM signals.

Why MultiMail

Built for Intercom developers

Approval gating before delivery

Intercom's native outbound email has no per-message approval step for AI-authored content. MultiMail's `gated_send` mode holds every agent-drafted message in a review queue until a human approves it via the API or dashboard — then delivers it with a cryptographic signature proving the approval happened.

Cryptographic sender identity

Messages sent through MultiMail carry a DKIM signature tied to your verified domain and an `X-MultiMail-Identity` header — an ECDSA P-256 signed claim covering the operator and mailbox identity, oversight mode, and capabilities — identifying the agent that authored the message. Intercom's shared mailbox provides neither. This matters for CAN-SPAM compliance and for recipients who want to verify the message came from a real, accountable source.

Immutable decision log

Every `POST /v1/oversight/decide` call — approve or reject — is written to an append-only audit log with timestamp, approver identity, and the original draft. If a regulator, customer, or internal reviewer asks what your AI sent and who authorized it, you have a complete answer.

Separate mailbox per agent role

Use `POST /v1/mailboxes` to provision dedicated addresses like `support-bot@yourdomain.com` or `onboarding@yourdomain.com`. Replies route back through MultiMail's inbound pipeline, keeping agent-originated threads separate from your human support queue in Intercom.

Webhook-native architecture

Intercom's workflow automation fires webhooks on conversation events. MultiMail's inbound webhooks handle delivery status, approval events, and replies — so the two platforms compose without polling loops or manual reconciliation.

Try it

Try it with your agent

No code, no dashboard. Paste this to your AI agent — it connects MultiMail, creates an inbox, and builds the flow for you.

1. Get MultiMail ready for this Intercom AI agent: read https://multimail.dev/llms.txt, connect the MultiMail MCP server in the development environment your agent uses, create a free inbox, and set up a verified sending domain so outbound email uses an approved sender identity. 2. In Intercom, wire MultiMail through Intercom’s Custom Actions capability for Fin AI Agent. Add actions that your Intercom agent can call when a conversation needs email outside the Intercom thread, and map each action to the MultiMail MCP tool exposed in your agent runtime rather than inventing a separate email API. 3. Give the Intercom agent three capabilities only: check the MultiMail inbox for relevant inbound mail, draft a reply for developer review, and send or schedule an approved email from the verified sender. Keep the action descriptions narrow so Fin uses Intercom for conversational support and MultiMail only for external email with identity verification, approval gating, and audit history. 4. Send a test email: ask the Intercom agent to draft a short message from the verified sender to a test address, confirm it appears as a pending MultiMail draft, review the recipient, subject, body, and sender identity, then approve the send. 5. Run the agent in gated_send mode for launch. Do not allow autonomous sending yet: every outbound email must pause for developer review and approval before MultiMail sends or schedules it. Move to monitored or autonomous only after the test flow and audit trail are consistently correct.
Getting started

Step by step

1

Create a MultiMail account and get an API key

Sign up at multimail.dev and generate a live API key (mm_live_...) from the dashboard. Use a test key (mm_test_...) during development — test mode queues messages but never delivers them.

2

Provision a dedicated mailbox for your Intercom bot

Use POST /v1/mailboxes to give your Intercom AI agent its own address. This keeps bot-originated threads separate from your human support queue and routes replies back through MultiMail.

3

Set up the Intercom webhook to trigger MultiMail

In Intercom's Developer Hub, subscribe to conversation.resolved and conversation.created events. Point them at your backend handler, which validates the X-Hub-Signature header and calls MultiMail's POST /v1/mailboxes/{mailbox_id}/send endpoint with the bot-drafted content, sending from a mailbox set to gated_send.

4

Wire up the approval review flow

Call GET /v1/oversight/pending periodically (or on webhook trigger) to surface messages awaiting review. Your team approves or rejects via POST /v1/oversight/decide. On approval, MultiMail delivers the message and fires your outbound webhook with the delivery receipt.

5

Verify delivery and check the audit log

After approving a message, read it back with GET /v1/mailboxes/{mailbox_id}/emails/{email_id} to confirm its delivery status. The approval itself is recorded in the audit log (GET /v1/audit-log). Together these are your evidence trail for CAN-SPAM compliance and internal audits.

FAQ

Common questions

Does MultiMail replace Intercom's native email channel?
No. Intercom handles inbound conversation management, triage, and Fin AI responses. MultiMail handles outbound email that originates from an AI agent and needs approval gating or cryptographic identity — typically follow-up emails sent after a conversation closes. The two work in parallel; you do not migrate your Intercom setup to use MultiMail.
Can I use gated_send for some message types and autonomous for others?
Oversight mode is a property of the mailbox, not a per-message flag. Provision separate mailboxes for different risk profiles — e.g. a gated_send mailbox for agent-drafted renewal reminders and an autonomous mailbox for automated delivery receipts — and route each message type through the appropriate mailbox by its mailbox_id. Audit logging applies to every mailbox regardless of mode.
How do I route customer replies back into Intercom?
Configure a webhook_url on your MultiMail mailbox. When a customer replies, MultiMail fires your webhook with the inbound message payload. Your handler can then call Intercom's Conversations API to create a new reply on the original conversation thread, keeping context intact.
Does this integration require an Intercom app or OAuth?
No app installation is required. The integration uses Intercom's outbound webhook subscriptions (available on all plans) and Intercom's REST API for reading conversation data. You store your own Intercom access token and MultiMail API key in your backend environment — no OAuth flow between the two platforms.
How does CAN-SPAM compliance work for AI-authored messages?
CAN-SPAM requires commercial emails to identify the sender, include a physical address, and honor opt-outs. MultiMail's DKIM signing and X-MultiMail-Identity signed claim establish sender identity. Your outbound templates must still include the required physical address and unsubscribe mechanism. The audit log of every POST /v1/oversight/decide call provides the human-in-the-loop record required by some internal compliance policies, but it does not substitute for required CAN-SPAM content elements.
What happens if a reviewer rejects a message?
The message is permanently cancelled — it is never delivered. MultiMail writes the rejection to the audit log with the reviewer identity and timestamp. The original draft remains readable via GET /v1/mailboxes/{mailbox_id}/emails/{email_id} for post-mortem review. Correlate the rejection back to the Intercom conversation by keeping your own mapping of the returned email id to the conversation_id at send time.
Related integrations

Explore more

Salesforce

workflow · both

View guide →

HubSpot

workflow · both

View guide →

Zendesk

workflow · both

View guide →
Get started

The only agent email with a verifiable sender

Email infrastructure built for AI agents. Verifiable identity, graduated oversight, and a hosted MCP server. Formally verified in Lean 4.

Read the Docs →See Pricing