We stopped opening a campaign builder to start a campaign. These days the first draft of a client’s cold email programme gets written in a conversation with Claude, and the campaign that comes out the other side is paced by the same scheduler that has always paced it. Nothing about the sending changed. What changed is where the setup work happens.
This is the exact setup we run, written down so you can copy it.
What “running cold email from Claude” actually means
It does not mean Claude sends email. That is the part people get wrong, and it is worth being blunt about before anything else.
Claude talks to WarmySender — the outreach tool we run our clients’ campaigns on — over the Model Context Protocol, an open standard for letting an assistant call a service’s tools. When you ask Claude to build a campaign, it calls a tool that writes the campaign. When you ask it to launch, it calls a tool that hands that campaign to WarmySender’s scheduler. The scheduler is what sends, at the pace it was always going to send at.
WarmySender’s own agent page puts the boundary plainly: “Claude never sends a message itself and can never raise a limit,” and “Creating or launching a campaign only writes it and hands it to WarmySender’s scheduler.” The scheduler “paces every email, LinkedIn, Instagram and WhatsApp action inside safe limits.”
There is a second boundary that matters more than it sounds. Connecting and disconnecting mailboxes stays in the app. As the Claude cold email page says, that is deliberate — “a credential should never pass through an agent or a chat log.” We like this rule because it means the worst possible outcome of a confused conversation is a badly written campaign sitting in draft, not a mailbox handed to something we did not intend.
The setup, step by step
1. Connect Claude to your workspace
In Claude on the web or desktop, open Customize → Connectors → Add custom connector and paste:
https://warmysender.com/mcpThen approve the sign-in. There is no key to paste and no key to lose — you sign in to your own workspace in the browser and Claude gets access to that workspace only. WarmySender’s connect-your-agent guide has the same three lines for every other client.
2. If you live in the terminal, use Claude Code instead
Half our team never leaves a terminal. One command:
claude mcp add --transport http warmysender https://warmysender.com/mcpThen /mcp inside Claude Code, or claude mcp login warmysender, and sign in.
We also install the plugin, which bundles skills on top of the raw tools so Claude has opinions about campaign structure rather than just the ability to create one:
/plugin marketplace add fortuneflick/warmysender-claude-plugin
/plugin install warmysender@warmysenderThe repo is public at github.com/fortuneflick/warmysender-claude-plugin. It describes itself as the “Official WarmySender plugin for Claude Code & Cowork — outreach campaigns, email verification and warmup via MCP, with bundled skills.”
3. Check what you actually connected
First thing we ask, every time, on every client workspace:
List my mailboxes and their health, then list every campaign and tell me which ones sent nothing yesterday and why.
This is a read-only question and it is the single most useful one in the whole setup. It replaced a fifteen-minute click-through that we used to do by hand at the start of every day.
What we actually ask Claude to do
The cold email tool set is narrow on purpose. Claude can create a multi-step campaign, edit it, and set the sending window and daily limits. It can start, pause and resume. It can add prospects one at a time or in bulk, update their details, and enrol or unenrol them. It can work from saved templates instead of writing every email from scratch. It can read and add to the suppression list. And it can report on every campaign and explain why nothing went out today.
In practice our day looks like four kinds of request.
Building
Create a four-step campaign for the fintech list, sending Tuesday to Thursday between 9 and 4 in the prospect’s timezone, capped at 40 a day per mailbox. Use our “operations audit” template for step one.
What comes back is a campaign in the app that we then open and read. We have never once launched something an agent built without reading it. That is not distrust of the model, it is the same review we would give a junior’s draft.
Hygiene
Add every address on this list to the suppression list, then tell me which of my running campaigns had any of them enrolled.
Suppression is the least glamorous and most valuable thing on this list. Being able to do it in one sentence, across a whole workspace, means it actually gets done.
Verification before launch
WarmySender’s MCP server exposes email verification tools alongside the campaign ones, and verification shares the same allowance and pacing as everything else. We verify a client list before enrolling it rather than after the bounces arrive.
Explaining a quiet day
“Why did this campaign send nothing?” used to be a support ticket. Now it is a question, and the answer is usually mundane — a start date in the future, a sending window that has not opened in that timezone yet, a daily cap already met.
Where we still click buttons
Connecting a mailbox. Disconnecting one. Anything involving a credential. Raising a limit — which, to be clear, no agent can do at all; WarmySender’s guardrails hold “regardless of what your AI asks for.” And the final read of any campaign before it goes live.
We are not trying to remove ourselves from the loop. We are trying to remove the twenty minutes of clicking between deciding what to do and it being done.
FAQ
Can Claude send a cold email on my behalf?
No. It writes and launches campaigns; WarmySender’s scheduler does the sending, inside your daily caps and ramp.
Can Claude raise my daily sending limit if I ask it to?
No. There is no tool for it. Limits, ramps and domain caps are enforced by the scheduler no matter who or what is asking.
Do I need an API key?
Not for Claude or Claude Code — both sign in through the browser. Keys exist for headless setups, and you create them in Settings → API Keys.
Can Claude connect a new mailbox for me?
No, and that is intentional. Connecting and disconnecting accounts stays in the app so a credential never passes through an agent or a chat log.
Which channels does this cover?
Cold email, email warmup, LinkedIn, Instagram, WhatsApp and email verification, from the same connection.