Product
Six things the API does that a send endpoint does not.
Sending is table stakes.
What makes this usable by an agent is that the reply comes back, the credential carries its own limits, and the platform refuses the send that would get your domain reviewed. Each page below is one of those, with the real request and response shapes.
New here? Start with what an email API for AI agents has to do, or how sending and receiving work as one round trip. To see it end to end, read the use cases: a support agent and a follow-up agent, each with an example you can run.
/v1/emails /v1/threads
/v1/emails/batch /v1/approvals
/v1/domains /v1/broadcasts
/v1/suppressions /v1/automations
/v1/templates /v1/topics
/v1/webhook-endpoints
The pages
One page per claim. Every claim matches the code.
Simple
One API
Transactional, campaigns, automations, templates, contacts and inbound behind one key, with one suppression list and one event stream.
The round trip
Replies
Inbound mail joined to the message it answers on Message-ID, the quoted history stripped, and one query for what is waiting on you.
Limits on the credential
Guardrails
A daily cap, a recipient allowlist and a human approval hold, on the key an agent holds rather than the workspace it belongs to.
Human in the loop
Human approval
A key that drafts instead of sends. A person releases from a queue, once, with the schedule kept and a 72-hour expiry, so an agent can write mail without being able to send it.
Campaigns
Campaigns
Broadcasts previewed before they start, automations with three automatic exits, topics people opt down from, and templates that fail on a missing variable.
Deliverability
Deliverability
Two sending identities per domain, suppression with a scope, permanent bounces only, and a campaign refused before it can hurt your reputation.
Start free, or read the docs first.
Inbound is included. Priced per email sent, never per contact. See pricing, or add the MCP server to an agent and let it read the API itself.