AgentMail

Operator's take

Any agent that does real work eventually hits email: confirming an appointment, chasing an invoice, receiving a verification code, replying to a customer. Wiring that through Gmail is miserable in a specific way — legacy providers have no API for creating inboxes, charge a monthly seat per address, impose sending limits designed for humans, and wrap everything in OAuth. AgentMail's bet is that the inbox itself should be an API primitive: one call creates a working email address for an agent, the way you'd provision an API key, and pricing follows usage instead of seats. Their line — "it's not AI for your email, it's email for your AI" — is an accurate description of which product category this is, and it's the one most people haven't seen yet.

The receiving side is what separates it from the transactional senders you may already know. Incoming mail pushes to your agent in real time over webhooks or WebSockets; threads, labels, drafts, and attachments are first-class API objects; there's full-text search across every inbox in the account, automatic labeling from your own prompts, and structured data extraction from message bodies. An MCP server is included on every tier, which means Claude-style agents can operate an inbox as a tool without custom glue code. The free tier — 3 inboxes, 3,000 emails a month, no card — is enough to actually pilot an agent workflow, and the $20/month Developer tier adds custom domains.

Two honest caveats. First, this is developer infrastructure: there's a console, but no operator-facing workflow lives here — you need an agent framework, a developer, or an automation platform in front of it before anything useful happens. Second, AgentMail is a young, seed-stage company (Y Combinator-backed, $6M seed announced on its site) in a category that barely existed a year ago; it publishes a SOC 2 report and a status page, but you're still adopting early-stage infrastructure and should weigh the switching cost accordingly.

What it's good at

• Inboxes as an API primitive — create and manage email addresses programmatically in milliseconds, with API-key auth instead of OAuth flows, and no domain verification needed to get started.
• Two-way email, not just sending — webhooks and WebSockets push incoming mail to your agent in real time, with threads and replies preserved in context; most email APIs only cover the outbound half.
• Agent-native tooling — a bundled MCP server, automatic labeling driven by your own prompts, structured data extraction from unstructured emails, and full-text search across all inboxes.
• A usable free tier — 3 inboxes, 3,000 emails/month, and 3 GB of storage at $0 with no credit card; Developer at $20/month adds 10 inboxes and 10 custom domains, Startup at $200/month scales to 150 of each.
• Deliverability plumbing handled — DKIM/SPF/DMARC, suppression lists, optimized shared IPs (dedicated IPs at Enterprise), and standard IMAP/SMTP access — connect the inbox to Outlook, Thunderbird, or Apple Mail, or to systems that speak those protocols rather than REST.
• Security posture unusual for its age — SOC 2 report (Startup tier and up), signed webhooks, multi-factor auth, and a public status page.

What it's not

• Not email for your human team — there's no mail client for people; your staff's inboxes belong in Google Workspace or Microsoft 365, with AgentMail handling the agents.
• Not a cold-outreach or newsletter engine — daily send caps (100/day free, 1,000/day Developer, 15,000/day Startup) and the whole design point at conversational, transactional agent email; volume outreach belongs in purpose-built sending tools.
• Not no-code — REST API, SDKs, and webhooks are the product; without a developer or an agent/automation platform wired to it, an operator has nothing to click.
• Not battle-aged — a seed-stage vendor defining a new category; the product moves fast and long-term durability is unproven, which argues for keeping your agent's email logic portable.