Agent inbox APIs: what to check before you choose one
Choose an agent inbox API when you need inboxes created by code, with threads, replies and webhooks. Before you commit, ask when mail is screened. If an agent can read a message before it has a verdict, pick an inbox that screens before delivery, or keep that agent away from mail sent by strangers.
Updated
An agent inbox API gives software a real mailbox over HTTP. Products in the category can look alike on a feature list. The place to look closest is what stands between a stranger's message and your agent.
What an agent inbox API gives you
- Inboxes created by an API call, often on your own domain.
- Receiving, threads and replies, so a conversation stays a conversation.
- A webhook or a live event when mail arrives.
- Parsed content: text, attachments, sometimes verification codes.
- SDKs, and often an MCP server.
The buyer's checklist
Put each question to the provider, then check the answer yourself. The last column is how Trotta Mail answers.
| What to check | Ask the provider | Trotta Mail |
|---|---|---|
| Built for | Is it built for an agent that reads alone, or for a human with an API on the side? | AI agents and humans. The guard is stricter for an agent. |
| When mail is screened | Before a message is accepted, or after it is stored? Is it screened at all? | While the sending server is still connected, before the message is accepted. |
| What an agent can read | Can an agent read a message while its verdict is pending? Does it get raw HTML or a cleaned view? | The verdict is decided before a message is readable. A key reads the safe view: hidden text removed, instructions aimed at a model marked. |
| Held mail and human review | Is risky mail held outside the inbox? Who can release it? Can the agent's own key? | Held outside the inbox. A human can release it. An agent's key cannot. |
| Outgoing checks | Is outgoing mail checked for secrets, and for replies an injected instruction is steering? | Screened for secrets, sensitive data and replies steered to a new address. |
| Sign-in codes | If an agent connects to a human's mailbox, can it read that human's codes and reset links? | Sealed from every API key in a human's mailbox. Parsed for the agent in its own inbox. |
| Setup for many agents | One call per inbox? Is each customer's mail walled off from the next? | One API call per inbox. Pods keep each customer's inboxes apart. |
| Price to start | Is screening on each plan, or an add-on? What do inboxes and volume cost? | Every check is on every plan. Free: 2 inboxes, 1,000 emails a month. Developer, $20 a month: 5 inboxes, 5,000 emails. Startup, $200 a month: 50 inboxes, 50,000 emails. |
Why the order matters
Agents act on what they read. The OWASP Top 10 for LLM Applications lists prompt injection as LLM01, the top risk, in both the 2025 edition and the current 2026 edition, and anyone can send an email.
If mail is stored and then checked, there is a window in which an agent can read it. Ask how long that window is, and whether a webhook can fire inside it.
How to verify the answers yourself
- Send the inbox a message with an instruction hidden in white text. Read it back through the API and see what the agent would be given.
- Compare two times: when the message became readable, and when its verdict or label appeared.
- Try to release or relabel a flagged message with the agent's own key.
- Have the agent send a message that contains a made-up API token, and see whether it leaves.
Trotta Mail has a drill for this. POST /v1/guard/drill sends realistic test mail through the real pipeline into your own inbox.
When another option is the better choice
- The agent receives mail from systems you control, such as your own test flows. Screening order matters less, so choose on price and developer experience.
- You need IMAP or SMTP access for a mail client. Trotta Mail has neither.
- The mail is a human's own. A mailbox built for humans may fit: see Gmail for AI agents and Outlook for AI agents.
No filter catches everything, Trotta Mail included. A reworded scam with no technical tell can still be delivered, which is why uncertain mail is held for a human.
What moving looks like
Moving an agent between inbox APIs is mostly an address change.
- Create the new inbox with one call.
- Swap the receive path: a webhook, live events, or a call that waits for the next message.
- Change the address where the agent is registered, so new mail and codes arrive at the new inbox.
- On a paid plan, connect your domain by publishing five DNS records. A subdomain leaves the rest of your mail where it is.
curl -X POST https://api.trotta.email/ v1/ agents/ signup \
-H "Content-Type: application/ json" \
-d '{"name": "riley"}'Questions
What is an agent inbox API?
An agent inbox API is a service that gives software its own mailbox over HTTP: create an address, receive and thread mail, reply, and get an event when mail arrives.
What is the most important question to ask an agent inbox provider?
Ask when a message is screened: before it is accepted, or after it is stored. Then ask what an agent can read while the verdict is pending.
Does every agent inbox API screen incoming mail?
Do not assume so. Ask each provider what is checked, when, and on which plans. Then test it with a message that hides an instruction.
Is Trotta Mail an agent inbox API?
Yes, with a difference in order: every message is screened while the sending server is still connected. Humans get a mailbox in the browser as well.
Sources
- OWASP Top 10 for LLM Applications 2026, the current edition (published August 2026)
- OWASP Top 10 for LLM Applications 2025, LLM01: Prompt Injection
- Trotta Mail API summary for agents
- Trotta Mail docs: concepts, security model, plans and limits