Claude Code Cold Email Automation: A Practical Setup Guide for B2B Teams

Jorge Macias

May 12, 2026

Table of Contents

TL;DR

Claude Code is Anthropic's terminal-based coding agent. With the right setup, it can run a full cold email pipeline for you: pulling leads, preparing mailboxes, kicking off warmup, drafting copy, and launching sequences. The trick is connecting it to your tools through MCP servers so you skip the curl-and-CSV grind. This guide walks you through the setup step by step.

What is Claude Code?

Claude Code is a coding agent that lives in your terminal. You type a prompt, it reads your files, runs commands, edits code, and tells you what it did. It works on Mac, Linux, and Windows.

What makes it useful for cold email is that Claude Code can run shell commands and write scripts. So if you give it API keys and tell it what to do, it can pull leads, hit endpoints, and check results. You do not write the code. You just describe the job.

The other piece that matters is the Model Context Protocol, or MCP. MCP is a standard way for tools to talk to AI agents. If a tool exposes an MCP server, Claude Code can call its actions without you writing a custom integration.

Why does this matter for cold email teams?

Cold email is a stack problem. To send one good campaign, you need leads, domains, mailboxes, warmup, copy, sequences, replies, and reporting. Most teams use four to six different tools to cover all of that. Each tool has its own dashboard and its own export.

When I used to run an agency project, I would spend half my day moving CSVs around. Pulling a list from one tool, cleaning it in a sheet, uploading it to another tool. Claude Code lets me skip most of that. I write one prompt and it handles the steps.

Here is what the manual way looks like next to the Claude Code way:

Step

Manual way

With Claude Code

Find leads

Open your lead tool, filter, export CSV, clean it

Ask Claude to pull leads using your ICP filters

Set up mailboxes

Buy domains, set DNS, create mailboxes one by one

Ask Claude to spin up domains and mailboxes via your infra provider

Warm mailboxes

Click around your warmup tool, watch heat scores

Ask Claude to start warmup and pull heat score reports daily

Write copy

Write each email by hand or in another tool

Claude drafts variants you review in your terminal

Launch sequence

Upload CSV, build sequence, click send

Claude pushes the sequence into your sender with one prompt

How do I set up Claude Code for cold email automation?

There are five steps. I will walk through each one in plain language.

Step 1: Install Claude Code

Open your terminal. Install Claude Code using the method that fits your machine. Anthropic supports Homebrew on Mac, WinGet on Windows, and apt or dnf on Linux. Once installed, run the claude command in any folder and log in. That is it.

Step 2: Get API keys from the tools in your stack

Sign up for the tools you need. At minimum, you need a sender, a lead source, a mailbox infrastructure provider, and a warmup tool. Each one gives you an API key in its dashboard. Copy each key into a safe place. You will use these to authenticate Claude Code with each tool.

Step 3: Connect MCP servers to Claude Code

This is the step that makes the whole pipeline feel native. If your tools offer MCP servers, you connect them to Claude Code once and it can call their actions any time. Most tools authenticate the MCP server with API keys passed as headers.

Add each MCP server to your Claude Code config. Run a quick test prompt like "List my active campaigns" and check that it returns real data. If it does, you are wired up.

If a tool in your stack does not have an MCP server, that is fine. Claude Code can still call its REST API through Bash and Python scripts. MCP just makes things cleaner.

Step 4: Build the lead-to-send pipeline

Now you tell Claude Code what to do. The prompt does not need to be long. Mine looks like this:

I want to send a cold email campaign to SaaS founders in the US.

Pull 200 leads filtered by job title "founder" or "CEO"

and industry "SaaS". Then create the campaign in my sender

using my warmed mailboxes. Use the email copy in copy.md.

Set follow-ups for day 3, day 7, day 12, and day 18.

Last follow-up is the breakup meme.

Claude reads my copy.md file, calls the right endpoints, builds the sequence, and reports back. It asks me to confirm before pushing the campaign live, which is the safety check I want.

Step 5: Add monitoring and replies

After the campaign is live, you want to know what is happening. Claude Code can pull reply data each morning and write a short summary. I have it write to a file called daily-report.md so I can scan it over coffee.

For replies, I still handle most by hand because the work is closing, not sending. But Claude Code can flag positive replies and queue follow-ups for me to review.

What does a real workflow look like day to day?

I open my terminal in the morning and run claude. I type a prompt like "Pull yesterday's reply data and group by sentiment." Claude pulls the data and writes me a markdown summary. I read it. I respond to positive replies. Done in 15 minutes.

Once a week, I ask Claude to check heat scores across all my mailboxes. Anything below 85, it flags. If a mailbox is dropping, I rotate it out and Claude updates the sequence.

Once a month, I ask Claude to pull a fresh batch of leads based on my ICP. The whole thing takes 20 minutes instead of half a day.

What mistakes should I avoid?

I made these mistakes so you do not have to.

  1. Skipping warmup. New mailboxes need around 14 days of warmup before real campaigns. Skip this and your domain reputation tanks fast.

  2. Letting Claude run unchecked. Always review the campaign before it goes live. Use plan mode or ask Claude to show the diff before changes.

  3. Using one sender for too long. Rotate mailboxes. Spread your volume so no single mailbox burns out.

  4. Running campaigns without follow-ups. A solid email sequence is at least 4 to 5 steps. Add LinkedIn touches if you can. Reply rate from follow-ups is almost always higher than the first send.

Which stack do I actually use?

People keep asking me what tools I plug into Claude Code for this. So here is the stack I have settled on. I have tried plenty of combinations and this one is the one I keep coming back to because every product exposes an MCP server, so Claude Code talks to them without custom glue.

It is called the Forge Stack from Salesforge. Six products, one MCP endpoint at mcp.salesforge.ai/mcp, and each piece handles one part of the pipeline.

Salesforge is the sender. Email and LinkedIn sequences from one dashboard. This is where Claude Code pushes campaigns.

Leadsforge is the lead source. 500M+ contacts with ICP search and waterfall enrichment. Claude pulls lists from here.

Mailforge handles cold email infrastructure on shared IPs. Domains and mailboxes ready in about 5 minutes.

Infraforge is the dedicated IP version. I use this when I am running high volume and want full control over sender reputation.

Primeforge gives me real Google Workspace and Microsoft 365 mailboxes for cold outreach. Ready in 30 minutes.

Warmforge handles warmup and deliverability monitoring. The recommended warmup is 14 days, and I aim for a heat score of 85 or higher before sending real volume.

You do not have to use all six. I started with just Salesforge and Warmforge, then added the rest as I scaled. The reason this stack works well with Claude Code specifically is the single MCP endpoint. One connection, six products, zero custom integration work.

FAQs

What is Claude Code?

Claude Code is an agentic coding tool from Anthropic that runs in your terminal. It reads your files, runs commands, and edits code based on plain English instructions. For cold email, it can run scripts, call APIs, and orchestrate tools through MCP.

Can Claude Code send cold emails on its own?

No. Claude Code writes scripts and uses MCP servers to talk to sending tools. The actual email delivery happens through your sending platform.

What is MCP?

MCP stands for Model Context Protocol. It is a standard way for tools to expose their functions to AI agents like Claude Code. If a tool offers an MCP server, Claude Code can call its actions without custom integration work.

Do I need to be a developer to use Claude Code for cold email?

You do not need to write code yourself. You do need to be comfortable opening a terminal, installing the Claude Code CLI, and managing API keys. After that, you describe what you want in plain English.

How is Claude Code different from no-code tools like Zapier or Make?

Zapier and Make connect tools through preset triggers and actions. Claude Code writes custom scripts on the fly and adapts to changes you describe in plain English. The trade-off is more responsibility for reviewing what it does before it runs.

How long does it take to set up cold email automation with Claude Code?

If you already have your sending tools and API keys, a basic pipeline takes about an hour to wire up. Warming new mailboxes takes around 14 days before you start sending real campaigns.

Final thoughts

Cold email used to feel like a job of clicking. Now it feels like a job of thinking. I think about the offer, the audience, and the timing. Claude Code handles the rest.

If you are just starting out, pick one tool you already use, find out if it has an MCP server, and connect it. Then add the next one. Build the pipeline piece by piece. You do not need to automate everything on day one.