Transactional email API.

AgentiSend is a transactional email API for applications and AI agents: verify a domain, create a key with a budget, send over REST or MCP, and read a log that says what happened to every message. Every key carries a send budget, a send the loop guard flags waits for a person, and one request pauses every sender. The docs are at https://agentisend.com/docs and the API contract at https://agentisend.com/openapi.json.

Every sender you set up has a sending limit and a speed limit. By default the limit counts per month and matches the emails your plan includes; you can lower it, or count it per hour or per day instead. We check it as each email goes out, so even a bot firing thousands of requests at once cannot slip past the number you set. When it hits the limit, sending stops and the bot is told exactly how to ask for more. A send the loop guard flags, or one an agent asks about first with request_approval, waits for a person. Approval rules you define yourself are not available yet. The same checks, written as the checklist people usually build themselves, are at https://agentisend.com/agent-email-guardrails.

No. Every automatic step is published: we warn you at 2% bounces / 0.05% complaints and pause at 4% / 0.08%, each by email with the numbers. If the rate keeps climbing, the pause can come soon after the warning. A paused account starts sending again after you appeal and a person reviews it. Abuse of the acceptable-use policy is stopped first and explained right after.

No. AgentiSend is for email people are expecting from you — receipts, alerts, resets, invoices. Sending to strangers who never asked breaks our rules and gets accounts paused, and we deliberately do not build the features that would make it easier.

Sending stops cleanly. Nothing goes out half-finished — each email is either accepted or refused, never something in between. The refusal tells your software how many it has sent, what the limit was, and how to carry on once the limit resets or you raise it.

One at a time. If you send to 500 people and one address is malformed, the other 499 go out and that one comes back with the reason. You never lose a whole batch to a single bad address.

You pick the part of the world your email is handled in, per domain, when you set it up — we never quietly choose for you.

Free keeps 7 days, Starter 30, Pro 90, Scale 180. Searchable from your account. Export before the window ends. We do not currently sell a longer add-on.

There is a written guide, not a wizard: it covers re-checking your domains, bringing across do-not-send lists, and where our API matches theirs. Most apps need only small changes. Where we are stricter, we warn rather than break. Plan for an afternoon, not a project. The guide is at /docs/guides/migrating-from-resend.

In the United States. Each sending domain records a region, US or EU; today mail for both is sent from our US nodes, and EU domains move to EU sending automatically when it ships. Account data and message history live with the product in one place; they are not copied to a second site for failover.

Yes — a public status page, with uptime figures measured once a minute by the same probe that writes the status history. Written uptime guarantees and service credits come with the Scale plan.

Yes. Every account starts on the Free plan: 5,000 emails a month, up to 170 a day, on 3 verified domains, with no card and no expiry. Paid plans start at $9 a month and have no daily cap. Every tier is at https://agentisend.com/pricing.

Yes. It is at https://api.agentisend.com/mcp over streamable HTTP, with a bearer key or OAuth 2.1, and it exposes the same operations as the REST API under the same limits and the same budget. One-line install for Claude Code, Cursor, Codex, Gemini CLI, VS Code, Windsurf and OpenCode is at https://agentisend.com/docs/guides/mcp.

Any language that can make an HTTPS request; there is a Node SDK on npm, and the quickstart shows Node, Python and curl. Guides with code that is run by the test suite against a local AgentiSend server exist for Next.js, Supabase, Better Auth, Auth.js, Hono, Express and SvelteKit, starting at https://agentisend.com/docs/guides/quickstart.

Resend is the better-known API, with React Email and a tier that costs nothing. AgentiSend adds the control plane an autonomous sender needs: a budget on every key, an approval queue, an account-wide kill switch, idempotency on every mutating endpoint, per-item batch results and an enforcement ladder published with its numbers. The sourced comparison is at https://agentisend.com/vs/resend.

Yes, to people who asked for them: broadcasts to a segment, topics each contact can opt out of, and one-click unsubscribe headers are part of the API. Cold outreach is not, and the product refuses sends that look like it.

One request. POST /limits/kill-all pauses every key on the account, every send then answers kill_switch_active, and POST /limits/resume-all turns sending back on; both write an audit row. To stop one key, POST /limits/keys/{id}/kill.

A code, a message, a fix naming the call that repairs it, and a docs_url, on every refusal. An agent that spends its budget reads agent_budget_exceeded with the period and the endpoint that raises the ceiling, and the key that asked cannot raise its own.

Nothing goes out twice. Every mutating endpoint accepts an Idempotency-Key header and remembers it, so a retry after a timeout replays the first result instead of mailing the same person again. Derive the key from the thing being done, never from the moment.