How to Send an Email Campaign from Claude (Step by Step)
Run a Claude email campaign end to end: connect Claude to AgentsMail, draft and render the template, test, approve the send, read the stats. Prompts included.

How to send an email campaign from Claude, step by step
To run a Claude email campaign, Claude needs a tool that can actually send: on its own it writes emails, it does not deliver them. Connect Claude to AgentsMail with the AgentsMail skill and an API key, and Claude can check your account, write and render the template, create the campaign, send you a test, and — once you approve — send it and read the results. This tutorial goes through the nine steps, with the prompt you give Claude and the API call it makes behind it.Can Claude send emails by itself?
No. Claude is a language model: it can draft a perfect campaign, but delivering it to 5,000 inboxes requires a sending service, a verified domain and an unsubscribe system. "Claude send email" in practice means: Claude calls an email platform's API on your behalf. That is what AgentsMail is built for. Its API v1 covers lists, contacts, templates, campaigns, sequences and statistics, and the AgentsMail skill teaches Claude how to use it — the endpoints, the shape of every answer, and the rules, starting with the most important one: never send without your explicit approval.| Task | Claude does it alone | Needs your approval |
|---|---|---|
| Check whether the account can send | ✅ | |
| Find or create a list, import contacts | ✅ | |
| Write and render a template | ✅ | |
| Create a draft campaign | ✅ | |
| Send a test to your inbox | ⚠️ it is a real send — Claude asks first | ✅ |
| Send or schedule the campaign | ❌ | ✅ every single time |
| Read stats, compare campaigns | ✅ |
Step 1: install the skill and give Claude a key
Run one command in your terminal. It asks which agent to install into — Claude Code, Cursor, Codex and others:bash
npx skills add https://github.com/mikecodeur/skills --skill agentsmailStep 2: ask Claude whether you can send
The first call is a diagnostic, not a campaign. It answers "can I send, and if not, why".text
Check my AgentsMail account: can I send a campaign right now?
If not, tell me exactly what is missing.ready is false, fix the blocker before writing a word. The most common one on a new account is domain_not_verified: Claude can declare your domain and fetch the DNS records to publish, and you paste them at your DNS provider.
Step 3: pick the audience
Ask Claude to list your lists with their contact counts, then choose one. If you are starting from a CSV, Claude can import it — up to 500 contacts per call, with a report of created, updated and rejected rows.text
List my AgentsMail lists with their contact counts.
I want to send to the "Newsletter" list.GET /api/v1/lists.
Step 4: write the template
This is where Claude shines. Give it the brief, the tone and one constraint per line:text
Write the November newsletter template: three short sections
(new feature, customer story, upcoming webinar), one button,
plain and readable on mobile. Use {{firstName}} but make the
greeting work when it is empty. Mark each section as an editable
zone so I can rewrite the copy later without touching the HTML.POST /api/v1/templates. Two rules it follows from the skill: the {{unsubscribeUrl}} link is mandatory, and each block a human should be able to edit carries a data-editable attribute. With those markers, you get a visual editor on the copy instead of a raw HTML box.
Step 5: render it, as many times as needed
Rendering is the only call with no side effect at all. Claude can iterate freely.text
Render the template and tell me about any warning or unknown
variable. Then show me the plain-text version.missing_postal_address: the render does not inject your postal address, the real send does. Claude knows to check your settings instead of "fixing" the template.
Step 6: create the draft campaign
text
Create a draft campaign "November newsletter" from that template,
to the Newsletter list, subject "What shipped in November",
with a preview text under 90 characters.Step 7: test it in a real inbox
A test send is a real email: it goes through the same checks and consumes one unit. Claude asks before sending it, then callsPOST /api/v1/campaigns/{id}/test-send with your address. Open it on your phone, in dark mode too. Tests leave no trace in the statistics.
Step 8: send or schedule
Once you approve, Claude calls the send endpoint — either now, or at a time you choose, written with an explicit time zone:bash
curl -s -X POST -H "x-api-key: $AGENTMAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{"scheduledAt":"2026-11-12T09:00:00+01:00"}' \
"https://www.agentsmail.io/api/v1/campaigns/<id>/send"202 with send_requested or scheduled. If something blocks the send — domain not verified, not enough sending units — the campaign is refused with a named reason and stays untouched: no half-sent campaign. Note that sending through the API is not open on every account yet; if it is not on yours, Claude prepares everything and you click Send in the app.
Step 9: read the results
text
How did the November newsletter do? Compare its click rate with
the October one and tell me which link got the most clicks.GET /api/v1/campaigns/{id}/stats: delivered, bounces, unique opens and clicks, unsubscribes, clicks per link. Rates are computed on delivered emails. Treat opens as indicative — privacy features preload images — and judge on clicks.
Prompts that work well with Claude
Before you create anything, tell me your plan in five lines and wait for my go.
Never send, test-send or schedule without asking me first, even if I approved a previous one.
If a call returns an error, quote the error and the blocker before trying anything else.These three lines remove most surprises. For the broader approach — why an agent-first email tool changes the workflow — read our guide to email marketing for AI agents, and our walkthrough of driving campaigns with AI agents. If you run a product, the same steps apply to email marketing for SaaS onboarding sequences.
FAQ
Can Claude send emails directly?
Not by itself. Claude needs a connected tool that sends email. With the AgentsMail skill and an API key, Claude calls the AgentsMail API to create and send campaigns, always after your approval for each send.Does it work in the Claude chat app or only in Claude Code?
The skill installs into agents that run commands, such as Claude Code. What matters is that Claude can make HTTP calls with your API key; any environment where it can do that can drive AgentsMail.Is it safe to let Claude manage my email list?
The risky action is the send, and it is gated: Claude must ask for approval for each campaign. Before that, AgentsMail checks your domain, sender and balance and refuses with a named reason rather than sending partially. Rendering has no side effects at all.How much does it cost?
AgentsMail's Free plan includes 3,000 emails a month and 10,000 contacts with the full REST API. Pro is €19 a month for 50,000 emails. API calls made by Claude are never billed separately.01Install the skill, add an API key, and start with the readiness check.
02Let Claude write and render freely — rendering sends nothing.
03The campaign copies the template: fix the campaign, not the template.
04Every test and every send needs your explicit go, one at a time.