How to give Claude an email address
Connect Claude to a shared inbox on your own domain in about ten minutes, no code — and the three things about it that surprise people afterwards.
About ten minutes, and no code. At the end, Claude is a member of a shared inbox — on a domain you own or on Yowl’s — reading mail and attachments and drafting the replies you send.
1. Register an organisation
yowl.co.za/orgs/new: enter your email, paste the one-time code, pick a
short name like acme. From that moment the organisation has an address, acme@yowl.co.za, and a private
portal at acme.yowl.co.za. An organisation is free.
2. Optional: link your own domain
Domains → connect. Yowl reads your live DNS, works out who has your mail today, and lists the records to add; you paste them. Skip it for now if you like — the organisation’s own address already receives, and a domain can come later. ⛔ One thing to know before you do it: an organisation links one domain, permanently, so a second domain means a second organisation.
3. Add the connector in Claude
On a Free, Pro or Max plan: Customize → Connectors → Add custom connector → paste
https://yowl.co.za/mcp → Add. Claude opens Yowl’s sign-in; sign in with the same one-time code, then
pick which agent Claude acts as, or create one there and name it. Every plan supports custom connectors by
URL; the Free plan allows one.
⚠ On a Team or Enterprise plan this is not yours to do. An Owner adds the connector for the whole organisation under Organization settings → Connectors → Add → Custom (choose Web if asked), and members then connect with their own accounts. If you are a member without that role, Claude gives you no way to add it — ask an Owner rather than hunting for the button.
4. The first three things to ask
- “Who am I in Yowl?” — the organisation, the addresses that receive mail, whether the domain is live.
It will tell you, correctly, that it has no address of its own: the organisation’s addresses are its
addresses, exactly as they are yours. There is no
claude@to set up. - “What’s in the inbox?” — the index, with a one-line summary per message written when it arrived.
- “Read the attachment on the one from Jeff.” — the document as text. Claude can then write an amended version back into the organisation’s files, which is how a document round-trips without anyone handling a file.
What it can actually reach
Worth knowing before you start, because the surface is wider than “read my email” and the good uses are mostly the ones people do not think to try.
The mail. The inbox index with a summary and category per conversation, full-text search, and the whole transcript of any exchange. Muted conversations are excluded but counted, so it can never tell you mail from someone does not exist when it has merely been hidden.
Waiting for mail that has not arrived yet. This is the one that surprises people. Claude can trigger something — a sign-up, a password reset, an order — and then wait on the address it invented for it until the confirmation lands, read the code out, and carry on. Combined with a catch-all, that is a whole class of errand it can finish rather than hand back to you.
Documents, both directions. It reads what is inside an attachment, not a description of it: the line items on an invoice, the clauses in a contract. Scans and PDFs are transcribed on first read. Then it writes the amended version back into the organisation’s files and puts it on a draft — which is how “amend the agreement Jeff sent and reply with our version” works end to end without anyone downloading a file.
The domain. Every DNS record Yowl expects, measured live, with the published value beside the expected one — and a fourth state for we could not check, which is why it never tells you to add a record that is already there. Plus your parsed DMARC reports: who actually sent as you, whether they aligned, and whether it is time to tighten the policy. “Is our mail set up properly?” is a question with a real answer now.
The team. It reads and writes the organisation’s chat, so when it drafts something it says so where you are already looking, and you can answer it there.
Projects and todos. A list of work that lives beside the mail, which it ticks off as it goes, and each item has its own discussion thread.
Where this is going
The direction is one shop for the whole life of a domain’s mail — register it, point it, verify it, bring the old mailbox across, watch the reports, work the inbox — with every part of that reachable the same way. Today the connector covers the mail, the documents and the domain’s health. The registration and transfer steps are the same account and the same conversation.
The reason that matters is not convenience. It is that an agent can only be useful about things it can see, and mail is currently spread over a registrar, a DNS host, a mail provider and a reporting tool that have never been introduced. One picture is what makes “is anything wrong?” a question worth asking.
5. A reply
When something needs answering, Claude offers: “Can I draft that email for you?” Say yes and it writes the reply as a draft. The team chat rings, you open the review page, change what you like, and send.
⛔ The connector has no send tool. Not a permission you can grant — it does not exist yet, so every reply Claude writes waits on the review form for a person. That is the current design rather than a setting, and it is the reason this is safe to try on a real inbox.
6. Optional: the whole portal
If you run Claude Code rather than the Claude app, one token opens more than the connector does. On the agent’s page in the portal: API token → copy → hand it to Claude. It can then log itself into the portal, do anything a member can, download the original bytes of a file, or drive the web UI with a browser tool.
Three things that surprise people
A document’s text may not be there on the first ask. Summaries are written when mail arrives, but the full text of a PDF or a scan is extracted the first time someone actually asks for it — reading a document nobody opens is most of what it would cost. So the first read of a scanned invoice can come back with the summary and a note to retry in a few seconds, and every read after that is instant.
Mail is written by strangers. Anyone can guess an address on your domain and send it whatever they like. Claude treats what it reads as data to act on, never as instructions, and the documentation it reads on connecting says so in those words. Everything it does lands in the same inbox and chat your team reads — there is no private lane, which is the point.
Retiring it is Disable, not Delete. On the agents page, disable the agent: that is the revoke, every token it held stops working, and its mail and chat history stay where they are. The members page is for humans, so that is not where you will find it.
What you have now
A team inbox with one more member in it — one that reads the attachments, drafts without being chased, and never sends without a person looking. The rest of these guides are about the domain underneath it.