← All problems and solutions

Can an AI agent create its own email API account?

Technically yes, if a service offers an unauthenticated signup call that returns an API key. But sending email also needs things only a person can supply: payment, ownership of a domain and its DNS records, and accountability for what is sent. Let the agent create the account and key, and keep billing, domain ownership and the go-live decision with a human.

By AISend. Updated . Also as markdown.

Why it happens

Account creation is the cheapest step in the chain, so it is the one that can be automated without argument. Any service with a signup API can hand an agent a credential in one request. The hard parts of email are elsewhere: a sender has to prove it owns the domain it sends from by publishing DNS records, has to pay for sending capacity, and has to answer for abuse and complaints. None of those are things an agent can honestly do on its own: it cannot own a domain or take responsibility.

That is why unattended signup is risky if it is the whole story. An agent that can both create accounts and send mail can be used to churn out throwaway accounts for spam. Providers therefore gate sending behind steps that need a human or a verified resource, even when signup is automated. If an agent tells you it "set up email" for you, check which steps it actually did and which are still waiting on you.

How to let an agent provision email safely (any provider)

  1. Separate creating an account from the authority to send. Decide that the agent may create credentials but not decide to go live. Going live means paying, pointing a domain, and choosing who gets mail. Keep those as explicit human actions in your process.
  2. List what a human must supply. At minimum: a payment method or plan, a domain you own with access to its DNS, a sender address on that domain, and acceptance of the provider's acceptable-use terms. Write the list into the agent's instructions so it stops and asks instead of improvising.
  3. Have the key go into a secret store, not the chat. A signup response contains a live key. Where the agent runs as a local MCP server, the key should be set in the client's environment config and the server restarted, rather than repeated in the conversation. Anything the agent prints is in its log.
  4. Let the agent do the technical prep. Adding a domain, fetching the DNS records to publish and polling for verification are API calls an agent can make. Publishing the records still needs whoever controls the DNS, so the agent should output the records and wait.
  5. Make the first send a supervised test. Send one message to an address you own, with the full message shown for approval, and check its delivery status before anything goes to customers. Use the new-account ramp-up limits that most providers apply as a natural safety margin.
  6. Name and log what was created. Record which agent created which account, when, and under which name. Give the key a descriptive name so you can find it in the provider's dashboard later.
  7. Plan the teardown. When the agent is retired, revoke its keys and remove the domain records. Unused credentials left behind are the usual source of later incidents.

What an agent can do alone on AISend and what needs a human

Taken from AISend's agent documentation and the API code. This applies to the local MCP server (npx -y aisend-mcp) and the REST API; the hosted MCP server does not offer account creation.

StepAgent alone?Detail
Create an account and API keyYesThe agent_signup tool or POST /api/v1/auth/agent-signup with a name (1 to 100 characters) returns an API key and a token without a browser. The endpoint is rate-limited per IP
Make the key active in the MCP clientNo, a config stepThe returned key takes effect only when set as AISEND_API_KEY in the client config and the server is restarted; the key appears in the conversation log
Add a sending domain and get DNS recordsYesPOST /api/v1/domains with the key returns the SPF and DKIM records to publish
Publish the DNS recordsNoWhoever owns the domain's DNS must do this
Verify the domainYes, after the DNS is livePOST /api/v1/domains/{domainId}/verify
Subscribe to a planNo, a human decidesAn account cannot send until a person subscribes to a plan; see pricing
Receive mail at the account addressNot possibleThe @agent.aisend.app address cannot receive mail and cannot log in to the dashboard
Send to customersOnly once the plan is active and the domain is verifiedA newly verified domain and a new account have lower daily caps at first

This table lists the steps; it is not a recommendation to automate every one of them.

When to use AISend

AISend supports the agent-first start: the local MCP tool agent_signup (or POST /api/v1/auth/agent-signup with {"name": "..."}) creates an account and returns an API key without a browser, email verification or password. The account cannot send yet. A person has to subscribe to a plan (AISend is paid; see pricing) and the account needs a verified sending domain you own. The automatic address ending @agent.aisend.app cannot receive mail and cannot sign in to the dashboard, so an agent account has no inbox of its own, and sending to anyone other than your own account email from the shared AISend address is refused. That gate is deliberate: it keeps throwaway signups from turning into spam.

Once the human has subscribed and the domain is verified, the agent operates through the nine hosted tools (send_email, get_email_status, list_emails, validate_email, list_domains, get_account_status, add_domain, verify_domain, list_api_keys). A key from agent signup is a full-access key, not the 90-day, 100-a-day OAuth connection that the hosted server issues, so treat it like any other key: store it safely and revoke it when you are done.

  • If you want the human owner to stay in the loop from the start, sign up in the dashboard and connect assistants with the OAuth sign-in instead of agent signup.
  • If you need an inbox the agent can read (replies, verification codes), AISend cannot do that: the agent address cannot receive mail. Use a mailbox or inbound-email service.
  • AISend is for transactional and consented mail; do not use automated signup to start cold outreach or bulk marketing.

Copy-paste prompt for an AI assistant connected to AISend

Connect an assistant first (setup for ChatGPT, Claude, Cursor and Gemini CLI), then paste:

Assign to AISend: I created an AISend account for you through AISend's agent signup. Check where we stand by calling list_domains and list_api_keys, then tell me whether a sending domain is verified and which keys exist. Do not send anything. If no domain is verified, show me what you did so I can approve the next step, and list the DNS records I would need to publish; do not ask me to paste any secret into this chat.

Frequently asked questions

Can the agent subscribe to a plan by itself?

It should not. A person should decide to subscribe, because that is a payment and a commitment to what the account will send. AISend's account cannot send until a human has subscribed and a domain is verified.

Can an agent account receive email?

No. The @agent.aisend.app address assigned at signup cannot receive mail or log in to the dashboard. Use a different service if the agent needs an inbox.

Is the hosted MCP server able to create accounts?

No. Account creation is available only through the local aisend-mcp server (agent_signup) or the REST endpoint. The hosted server only works for an existing account that a human has approved over OAuth.

What is a safe first send for a new agent account?

One message to an address you own, from the verified domain, with the exact message shown to you before it goes, followed by a status check. Then widen gradually.

Sources

Related