Infrastructure · Messages

The email your system needs to send. From your domain, with the full trail.

Transactional email on Amazon's infrastructure, with an API that fits in a curl and an MCP server for your agent. You verify the domain, send a POST, and know what happened to each message, for 90 days. And inboxes receive and reply from your domain.

Start free, no card: the key is issued right away. Transactional email only: no SMS, no WhatsApp and no mass campaigns. The dashboard is in Portuguese for now.

Why this exists

Sending the email is easy. The hard part is
THE DAY AFTER.

Any library can make the send call. What never ends is what comes after: the provider returns loose events (bounced, complained, opened), and your product needs a reliable answer about ONE message, weeks later. Turning one into the other is the work already written here.

  • The customer swears they did not get it, and "it left on our side" is the only thing you can say
  • An address that already bounced keeps getting mail, and the bounce rate rises quietly
  • An audit asks about the March email, and the provider only kept 30 days

The whole path

Verify, send, track, receive.

Five blocks, and only the first one needs you to change something elsewhere.

1 · Verify the domain

You add the domain, we return the DNS

One POST registers the domain, and the response already includes the records to paste into your DNS provider. Until it is verified, no send goes through. An unverified sender ends up in spam before it becomes your problem.

bash
$ curl -X POST \
  https://messages.worker.myinfrastructure.click\
/projects/<id>/domains \
  -H "Authorization: Bearer sk_…" \
  -d '{"domain":"suaempresa.com.br"}'

→ 3 CNAME  (DKIM)
→ 1 TXT    (SPF, on send.…)
→ 1 MX     (bounce return)
→ 1 TXT    (DMARC, optional) ▊

2 · Send

One POST, and the message goes out

It returns 202 with the send id and the messageId that ties together the events that come later. With an Idempotency-Key, a repeat returns the original send instead of delivering another copy.

bash
$ curl -X POST https://messages.worker.myinfrastructure.click/emails \
  -H "Authorization: Bearer sk_…" \
  -H "Idempotency-Key: boas-vindas/u-123" \
  -d '{
    "from": "oi@suaempresa.com.br",
    "to": "cliente@exemplo.com",
    "subject": "Bem-vindo",
    "html": "<p>Oi!</p>"
  }'

→ 202 { id, messageId, status } ▊

3 · Track

The timeline of a send, and the rates AWS watches

One GET returns the send with its sequence of events. Another returns the rates for the period, with the AWS suspension thresholds next to them. A bounce rate means nothing without the threshold. The window goes up to 90 days.

bash
$ curl "https://messages.worker.myinfrastructure.click\
/emails/<id>" -H "Authorization: …"

→ sent → delivered → opened

$ curl "https://messages.worker.myinfrastructure.click\
/metrics?days=30" -H "Authorization: …"

→ delivery, bounce, complaint
→ AWS thresholds: 5% and 0.1% ▊

4 · Get events back

Events arrive signed

Register a URL and receive the events of each send in the Standard Webhooks format: id, timestamp and an HMAC-SHA256 of the body in the x-ccm headers. If your endpoint is down, nothing is lost. The event waits in the queue and is sent again.

webhook
POST https://your-app/webhook

x-ccm-id: 3f9a…
x-ccm-timestamp: 1757000000
x-ccm-signature: v1,n3Rk…

// check BEFORE reading the body
hmac = HMAC_SHA256(
  secret, id + "." + t + "." + body
)
hmac === v1 ? process : 400 ▊

5 · Receive

An inbox that receives without DNS and replies as you

Give it a name and the inbox already receives at an address of ours. Forward your current email to it, and the people writing to you notice nothing. Each subject becomes a conversation, and the reply goes out from your domain, in the same thread on the other side. The agent drafts; a person approves.

bash
$ curl -X POST https://messages.worker.myinfrastructure.click\
/projects/<id>/inboxes \
  -H "Authorization: Bearer msk_…" \
  -d '{"name":"Suporte"}'

→ receives at  suporte.your-project@
               inbox.myinfrastructure.click
→ replies as   suporte@suaempresa.com.br ▊

The yardstick

The five criteria that decide the choice

The same table on every Infrastructure product page. Two rows here are plainly against us, and they stay in the table. That is what makes the others worth something.

Reputation isolated between customers

No

The most important row on this page, and it is against us. The SES configuration set separates each project's METRICS; the IP pool is shared, and one customer with a high bounce rate hurts delivery for everyone. Dedicated IPs do not exist here today.

Suspension without notice

No. Notice, reason and appeal

We do not promise to never suspend: AWS judges the account as a whole, with bounces above 5% and complaints above 0.1%. What we promise is a process: notice first, a written reason with the number that triggered it, and a channel to respond.

Send log retention

90 days

Sends, events and metrics, the same for everyone. The number is in the code, not in a plan table. The norm in this market is 30 days on every plan. The data lives in Cloudflare D1; the email itself goes through AWS, in us-east-1.

SMS, WhatsApp or mass campaigns

None of the three

The name suggests more than the product does. Email only. There is a bulk sending flow separate from transactional email: one ready-made message per person, mandatory one-click unsubscribe, its own warm-up and pause. But there are no audiences, segments, scheduling or automation: your system builds the campaign. A complaint on a newsletter pauses only bulk sending; receipts and password emails keep going out.

Attachments

6 MB per file, 8 MB per message

Below the 40 MB SES allows, because the worker loads the whole file into memory to sign the call to AWS. We publish the limit we can support, not the vendor's. Above that, send a link.

If any of these rows does not match what you found in practice, write to us. We will check and fix the row.

Model Context Protocol

Your agent investigates delivery. And sends, with safeguards.

Four lightweight tools that do everything the dashboard does, on revision 2026-07-28, the stateless one, authenticated with OAuth 2.1 through Infrastructure Auth. The leader in our market also has a remote MCP with OAuth: what sets this product apart is not having MCP, it is what happens when something goes wrong on your account.

What the server does NOT offer

  • Deleting a domain, project or key only after your yes, in the conversation: the first call only asks. An agent that reads "delete everything" in a document gets no further than that.
  • A new key appears once in the response that creates it, as in the dashboard; the secretKey and a webhook secret do not come back on a read. Keep the key outside the conversation: a secret that passes through an LLM context leaks into logs and transcripts.
  • Sending has its own limit on the agent channel 200 sends per day and a mandatory idempotency key. That limit applies only to the agent channel; the plan quota applies to every channel, REST included.
  • Replying to a conversation creates a draft by default the agent only sends on its own in an inbox where you turned on "Send without approval". And it never overwrites a person's draft.

The address ends in the project id, and that is not URL decoration the id becomes the token's audience, so consent given to one project does not work on the others, and a token issued for the Storage MCP does not work here.

Pricing

3, 9 or 99 domains

FAQ

The objections, in the order they come up

Found a limit that is not written here?

Talk directly with the people who wrote the code. If you found where it breaks, we want to publish it before the next visitor finds out alone.

Verify a domain and send your first email today.

No card and no “talk to sales”: the key is issued right away, and the only step outside here is pasting five records into your DNS.