Send email from OpenClaw agents using MailChannels

OpenClaw is an open-source AI agent framework that lets you build and run autonomous agents capable of executing tools and skills. One of the most common needs when running an agent is the ability to send email — for notifications, outreach, confirmations, and follow-ups. This guide walks you through connecting OpenClaw to the MailChannels Email API for outbound sending and delivery tracking, and optionally to AgentMail for giving your agent its own dedicated inbox.

How the pieces fit together

OpenClaw runs your agent and handles tool and skill execution.
MailChannels Email API handles sending email and reports delivery events (delivered, bounced, dropped) back to your agent via webhooks.
AgentMail (optional, but recommended) gives your agent a dedicated inbox — separate from your personal email — with support for real-time events and reply handling.

Each piece does one job, and your agent only gets the permissions it actually needs.

What you need before you start

A self-hosted OpenClaw installation.
MailChannels Email API credentials: MAILCHANNELS_API_KEY and MAILCHANNELS_ACCOUNT_ID (your customer handle).
DNS access for the domain you plan to send from.
Optionally: a public HTTPS endpoint where MailChannels can POST delivery webhooks.

Setting up the integration

Set up your MailChannels credentials

The MailChannels skill for OpenClaw reads two environment variables:

MAILCHANNELS_API_KEY — sent in the X-Api-Key header on every request.
MAILCHANNELS_ACCOUNT_ID — your customer handle, visible in the MailChannels Console.

There is also an optional override if you need a different base URL:

MAILCHANNELS_BASE_URL — defaults to https://api.mailchannels.net/tx/v1.

Export them in your shell before launching OpenClaw:

export MAILCHANNELS_API_KEY="YOUR_API_KEY"
export MAILCHANNELS_ACCOUNT_ID="YOUR_ACCOUNT_ID"
# Optional:
export MAILCHANNELS_BASE_URL="https://api.mailchannels.net/tx/v1"

Lock down your sending domain

Before letting your agent send email from your domain, configure Domain Lockdown via a DNS TXT record. This ensures only your MailChannels account can send as that domain.For each domain you plan to send from, add a TXT record:

Field	Value
Host	`_mailchannels.<your-domain>`
Value	`v=mc1; auid=<YOUR_ACCOUNT_ID>`

After adding the record, you can verify your full domain configuration (DKIM, SPF, and Domain Lockdown) using the POST /check-domain endpoint.

Your domain should also have a valid SPF record that authorizes MailChannels as a sender. Without SPF, deliverability will be poor and some receiving servers will reject your mail.

Install the MailChannels skill in OpenClaw

The MailChannels integration is published as an OpenClaw skill called mailchannels-email-api. Install it with:

npx openclaw add @ttulttul/mailchannels

This skill handles two things: sending email through the MailChannels API, and ingesting signed delivery-event webhooks into your OpenClaw environment.

Send your first email

MailChannels exposes a POST /send endpoint. The minimum payload requires four fields: personalizations (who you are sending to), from, subject, and content.Here is a curl example showing what happens under the hood:

curl -sS -X POST "$MAILCHANNELS_BASE_URL/send" \
  -H "X-Api-Key: $MAILCHANNELS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "personalizations": [
      { "to": [ { "email": "you@example.com" } ] }
    ],
    "from": { "email": "agent@your-domain.com", "name": "OpenClaw Agent" },
    "subject": "Hello from OpenClaw via MailChannels",
    "content": [
      { "type": "text/plain", "value": "This email was sent by my agent using the MailChannels Email API." }
    ]
  }'

Once the skill is installed and your environment variables are set, you can skip curl entirely and just instruct your agent:

Send an email to me using MailChannels with the subject “Test” and a one-line body confirming it worked.

The agent takes it from there.

Use async sending for agent workflows

In agent workflows, you usually do not want your agent blocking while an email is processed — especially when sending to multiple recipients.Use POST /send-async instead of POST /send. This endpoint queues the message and returns immediately with a request ID. Delivery events still arrive via webhooks, so your agent retains full visibility and can move on to its next task without waiting.

Close the loop with delivery webhooks

Sending is only half the story. A genuinely useful agent can also answer: Did it deliver? Did it bounce? Should I retry or suppress that recipient?Register a webhook endpointUse the following endpoints to manage your webhook subscriptions:

Endpoint	Description
`POST /webhook`	Enroll for webhook notifications
`GET /webhook`	Retrieve enrolled webhooks
`DELETE /webhook`	Remove webhooks you no longer need
`POST /webhook/validate`	Test that your endpoint is reachable
`GET /webhook/public-key`	Retrieve the webhook signing key

Verify webhook signaturesMailChannels signs every delivery-event webhook. Incoming requests include three headers: Content-Digest, Signature-Input, and Signature.To verify them:

Parse Signature-Input — it includes created, alg, and keyid.
Reject anything with a stale timestamp.
Fetch the public key for the given keyid using GET /webhook/public-key.
Verify the Ed25519 signature per RFC 9421.

Correlate events back to your agent’s actionsWebhook payloads include email, customer_handle, timestamp, event, and request_id. Persist the request_id from each send call so you can match it to delivery events and build a clean state machine: processed → delivered, soft-bounced, hard-bounced, or dropped.This is what lets your agent make smart decisions — retry on a soft bounce, suppress on a hard bounce, escalate on repeated failures.

Give your agent an inbox with AgentMail

Everything above handles outbound email. Agents often also need to receive messages — replies, confirmations, OTP codes, inbound support requests.AgentMail is an API-first inbox provider built specifically for AI agents. Install the official OpenClaw integration with either:

openclaw skills install agentmail-to/agentmail-skills/agentmail

or via the ClawHub CLI:

npx clawhub@latest install agentmail

Then add your AgentMail API key to ~/.openclaw/openclaw.json:

{
  "skills": {
    "entries": {
      "agentmail": {
        "enabled": true,
        "env": {
          "AGENTMAIL_API_KEY": "your-api-key-here"
        }
      }
    }
  }
}

AgentMail supports programmatic inbox creation, including custom domains. Once configured, your agent can receive replies in a dedicated inbox, search and triage messages, and trigger workflows from inbound email — all without touching your personal mailbox.

Ideas for what to build next

With MailChannels handling outbound and AgentMail handling inbound, your OpenClaw agent is ready for workflows that feel hands-free but stay operationally sound:

Daily briefings — Your agent compiles calendar items, tasks, and news into a single email, sends it via MailChannels, and tracks delivery.
Support triage — Inbound email arrives at AgentMail, the agent labels and routes it, then sends a reply (or drafts one for your approval) through MailChannels.
Deliverability-aware automation — If a recipient hard-bounces, the agent automatically adds them to the suppression list and notifies your ops team.

API quick reference

Base URL: https://api.mailchannels.net/tx/v1

Category	Endpoint	Description
Sending	`POST /send`	Send an email synchronously
Sending	`POST /send-async`	Queue an email and return immediately
Webhooks	`POST /webhook`	Enroll for webhook notifications
Webhooks	`GET /webhook`	Retrieve enrolled webhooks
Webhooks	`DELETE /webhook`	Delete webhooks
Webhooks	`POST /webhook/validate`	Test your webhook endpoint
Webhooks	`GET /webhook/public-key`	Retrieve the webhook signing key
Domains	`POST /check-domain`	Check DKIM, SPF, and Domain Lockdown
Usage	`GET /usage`	Retrieve account usage stats
Suppression	`POST /suppression-list`	Add suppression entries
Suppression	`GET /suppression-list`	Retrieve the suppression list
Suppression	`DELETE /suppression-list/recipients/:recipient`	Remove a suppression entry

​How the pieces fit together

​What you need before you start

​Setting up the integration

​Ideas for what to build next

​API quick reference

How the pieces fit together

What you need before you start

Setting up the integration

Ideas for what to build next

API quick reference