Sendmux: Email API for Products and AI Agents

Sendmux is one workspace and API to send, receive, route, and monitor email, built for modern products and AI agents. Most email services stop at sending. Sendmux gives your application or agent a real inbox that can send mail, receive it, hold it, reply, and search across threads, all through a single API and one dashboard.

If you have ever wired up one provider for outbound email and a second, clunkier setup for inbound, you know the pain. Sendmux puts sending and receiving in the same place, adds delivery monitoring on top, and treats AI agents as first-class users instead of an afterthought.

What Sendmux does

At its core, Sendmux is an email infrastructure platform with three surfaces: a Sending API for transactional email, a Mailbox API for reading and replying to inbound mail, and a Management API for domains, mailboxes, webhooks, logs, and billing. You can use one, two, or all three.

Send single or batch transactional email over a simple REST API. Spin up real mailboxes on your own verified domain or on the shared myagent.mx domain for quick tests. Read, triage, reply to, and sync inbound messages. Then watch every send land in a searchable delivery log you can export to CSV.

Built for AI agents

This is where Sendmux is different from a typical transactional email API. AI agents can have their own email on Sendmux, with a human kept firmly in the loop.

An agent can self-register for its own constrained mailbox through a challenge-first flow, then invite you, the owner, to approve it. Until you approve, the agent can read and receive mail but cannot send a single message. Human approval is the default, not a bolt-on setting. So you can hand an autonomous agent a real email address without handing over your domain's sending reputation.

For teams building with LLMs, this solves a real problem. An agent that can email customers, read the replies, and act on them is far more useful than one that can only fire off a message into the void. And because sending stays gated behind owner approval, you get that capability without the risk of an agent going rogue in your customers' inboxes.

Agent access and approval flow

The flow is designed to be safe by default. An agent discovers the protected resource, solves a proof-of-work challenge, and registers for an anonymous identity with a constrained mailbox. It saves its credential bundle, invites its human owner, and waits. Once you approve sending from the Agent access card on your team page, the agent can exchange its claim token for a sending-capable token. Before that, it simply cannot send.

Key features

Sendmux ships with the pieces you need to run email in production. Transactional sending covers single and batch email through the Sending API. Real mailboxes live on your verified sending and receiving domains, or on the shared myagent.mx domain when you just want to test something in two minutes. Full inbox access lets your code read, triage, reply, and keep messages in sync.

On the operations side, delivery logs and email metrics are searchable and exportable to CSV, so you can see sender, recipient, status, provider, attempts, and timestamps for every message. Signed webhooks push delivery and inbound events to a URL you control, with signature verification so you can trust what you process. API keys are scoped separately for sending, mailbox, and management work, which keeps a send-only key from ever touching account settings.

Each team starts with fifty active mailboxes, which is enough to give many agents or product features their own address.

How you connect

Sendmux meets you where you already work. There is a hosted MCP (Model Context Protocol) server with OAuth, so AI tools and agent frameworks can connect without you distributing raw API keys. There is a command-line interface for terminal workflows. And there are official SDKs for Python, Go, Ruby, PHP, and TypeScript, plus a plain REST API if you prefer to call it directly.

The hosted MCP option is worth calling out for anyone building agents. It keeps Sendmux in the authorization flow, uses short-lived tokens, and only exposes the tools allowed by the grant. That means an agent gets exactly the email access you intend and nothing more.

Use cases

Developers reach for Sendmux when they want transactional sending and a programmable inbox in one API, instead of stitching two vendors together. Product teams use it to send onboarding email, receive replies, and route inbound messages back into the app. Teams building AI agents use it to give each agent its own email address for outreach, support, scheduling, and follow-up, with owner approval guarding every send.

It also fits internal tooling, notification systems, customer support automation, and any workflow where a machine needs to both send and read email rather than just push it out.

Why Sendmux over a send-only email API

Traditional transactional email APIs are excellent at delivery and stop there. If your product needs to read replies, hold conversations, or let an agent act on what came back, you end up bolting on a separate inbound service and gluing the two together. Sendmux removes that seam. Sending, receiving, routing, and monitoring share one workspace, one set of scoped keys, and one delivery log.

For AI use cases specifically, the owner-approval model is the real draw. You get the upside of an agent with its own inbox and the safety of a human signing off before anything leaves the building.

Frequently asked questions

Can Sendmux receive email, not just send it? Yes. Every mailbox can send and receive, hold messages, and be read or replied to through the Mailbox API.

Is Sendmux good for AI agents? That is the core use case. Agents can self-register for a constrained mailbox and only send after a human owner approves them.

What languages and tools are supported? A REST API, a CLI, a hosted MCP server with OAuth, and SDKs for Python, Go, Ruby, PHP, and TypeScript.

Do I need my own domain? You can use a verified custom domain for production, or the shared myagent.mx domain to test in minutes.

How do I monitor deliveries? Searchable delivery logs, exportable metrics, and signed webhooks for delivery and inbound events.

Give an agent its own inbox, keep a human in the loop on every send, and see exactly what went out. That is Sendmux.