Agent Mail

Real email inboxes your AI agents own. Create them by API or from an MCP tool, receive and reply to mail, and let attachments land straight in an AI folder your assistant can already read and search. Runs on our own UK network - no third-party email service in the path.

Why Agent Mail

• Identity - a verifiable address (name@agents.cloudbox.biz, or your own domain) so an agent can sign up for services and receive their confirmations.
• File-native attachments - every attachment is stored in the inbox's AI folder, so read_file and search work on it seconds after it arrives.
• Memory - messages and threads are a durable, searchable record the agent can return to.
• UK-sovereign - inbound and outbound run on our own network, DKIM/SPF/DMARC aligned.

Two ways to use it

From an AI assistant (MCP)

Once your CloudBox connector is added to Claude, ChatGPT or any MCP client, the agent-mail tools appear automatically - no code:

create_email_inbox   { "name": "billing-bot" }
list_emails          { "inbox": "billing-bot", "unread_only": true }
read_email           { "message_id": "msg_..." }
send_email           { "inbox": "billing-bot", "to": "ap@acme.com", "subject": "Re: invoice", "text": "Paid, thanks." }
wait_for_email       { "inbox": "billing-bot", "match": "code", "timeout": 60 }

From code (REST)

# create an inbox
curl -X POST https://app.cloudbox.biz/api/v1/mail/inboxes \
  -H "Authorization: Bearer $CBS_KEY" -H "Content-Type: application/json" \
  -d '{"name":"billing-bot"}'

# wait for the next email (great for OTP / verification codes)
curl "https://app.cloudbox.biz/api/v1/mail/inboxes/inb_123/messages/wait?timeout=60&match=code" \
  -H "Authorization: Bearer $CBS_KEY"

Authentication uses your existing CloudBox API key - the same cbs_ secret as the Storage API. Full endpoint list in the API reference.

Anti-abuse

Agent Mail is for transactional agent workflows, not bulk or marketing mail. Every plan has a published daily send cap, a max recipients-per-message limit, duplicate-content blocking and automatic bounce suppression. See acceptable use.