The short version
- Mailchimp's own official MCP server covers transactional mail (Mandrill) and cannot build a campaign, so this uses a community server plus an API key.
- Add the server to Claude, paste your API key, and confirm it can list your audiences.
- Brief the campaign, let Claude draft it, then approve the copy before anything is scheduled.
Why Claude + Mailchimp is a fast email stack
Almost everyone has a Mailchimp account. It's the platform most people reach for first (newsletters, campaigns, audiences and tags, Customer Journeys), and it's huge with small businesses and creators. The catch is that the actual work is slow. Writing the issue, picking the segment, wrangling the subject line, formatting the blocks: it eats an afternoon every time. So newsletters go out late, or go stale, or quietly stop.
Claude changes the economics. Connected to your account, it can read who your subscribers are, plan who to talk to and why, and draft the whole thing (the segments, the campaign, the journey emails) in the time it used to take to settle on a subject line. The one thing it shouldn't do is press send on its own. That's the job this workflow is built around: Claude does the heavy lifting; you approve; Mailchimp ships.
The missing piece is somewhere for all of that to live. Drafts inside a chat window scroll away, and an agent firing changes straight into Mailchimp with no review is how you end up emailing the whole list twice. Plainpaper is the canvas in the middle: every segment, campaign, and email the agent produces lands as a typed card you can see, question, and approve, with the audience and the offer that shaped it attached.
What each tool does in this workflow
Each of these three tools has a clear lane, and the workflow is strongest when you let them stay in it.
The workflow, end to end
Six stages, each producing a persistent artifact that feeds the next. The whole thing runs from the chat window you already use. Claude does the work and writes cards onto the board in real time.
What you'd see: Claude calls Mailchimp over MCP, then drops the drafted newsletter onto the board. Nothing is scheduled yet.
How to connect Mailchimp to Claude
One thing to settle before you start, because most write-ups skip it. Mailchimp does not publish an MCP server for the Marketing API, which is the half of Mailchimp that holds your audiences, campaigns, and Customer Journeys. Its one official MCP server covers Mailchimp Transactional (Mandrill), the service that sends receipts and password resets, and it cannot build a campaign. So there is no "log in with Mailchimp" button here. You pick one of these routes, and the first three all end up calling the same Marketing API.
| Route | What you get | What it asks of you |
|---|---|---|
| A community MCP server Marketing API, runs on your machine | The full campaign surface as tools: create the draft, set the HTML, send a test, schedule, plus audiences, segments, tags, and reports. | You install it and hand it a Mailchimp API key. Nobody at Mailchimp reviews these. |
| A hosted connector Zapier MCP and similar | A smaller set of actions wired to your Mailchimp login, including create campaign, send campaign, and subscriber management. | Nothing to install, but a narrower surface than the API and one more vendor holding the connection. |
| The Marketing API itself straight from Claude Code | Every endpoint Mailchimp publishes, with no MCP server in the middle. Claude Code calls the REST API with the tools it already has. | You work in a terminal, and the key lives in your environment. |
| Mailchimp Transactional MCP the official one | About nine tools at mandrillapp.com/mcp for transactional sending, templates, and delivery troubleshooting. |
Worth knowing it exists so you don't reach for it here. It cannot create or send a marketing campaign. |
Doing it in Claude Code
Claude Code reaches Mailchimp either way. To add a community Marketing API server, which runs locally and picks the key up from its environment:
claude mcp add --env MAILCHIMP_API_KEY=abc123-us8 \ --transport stdio mailchimp -- uvx mailchimp-mcp
That example runs mailchimp-mcp, one of several community
servers on the MCP registries. Read what a server does before you give it a
key, and prefer the ones that ship a read-only and a dry-run mode.
Or skip MCP altogether. The Marketing API is plain REST with basic auth, so
Claude Code can call it directly. The data center suffix on your key (the
us8 in abc123-us8) is the subdomain you talk to:
curl --url 'https://us8.api.mailchimp.com/3.0/campaigns' \ --user "anystring:$MAILCHIMP_API_KEY"
A campaign is three calls: POST /campaigns for the draft with
its audience and subject line, PUT /campaigns/{id}/content for
the HTML, and POST /campaigns/{id}/actions/schedule once you
have approved it. Point Claude Code at the Marketing API reference once and
it will chain them itself.
About that key. A Mailchimp API key carries everything the login that created it can do. There are no scopes and no per-tool limits, so treat it like a password: generate it under Account, Extras, API keys, keep it out of shared chats, and delete it when you stop using it. If your server offers read-only or dry-run mode, run the first week with both on.
Step by step: a campaign from brief to send
Connect Claude to Mailchimp and a board
Everything here runs over MCP, the open protocol that lets Claude call other tools. The Plainpaper board is a remote server you log into. Mailchimp takes one of the routes above, so pick that first: a community MCP server, a hosted connector, or the Marketing API called straight from Claude Code. Once both are wired up, the same conversation can look up an audience, draft a campaign, and put it somewhere you can approve it.
Permissions matter. Let Claude read your audience freely, but keep Mailchimp's send and schedule actions behind your approval on the board. The agent can draft a campaign on its own; it can't ship one without you.
Brief Claude like you'd brief a marketer
Generic AI copy reads generic because the model is guessing at the point. Don't make it guess. Tell Claude the goal, the offer, and who you're talking to, the same three things you'd hand a person. That short brief is the difference between "Check out our updates" and an email that actually has a reason to exist.
Goal: get lapsed subscribers opening again. Offer: 20% off their next order, 7 days only. Audience: no opens in 90 days but bought before. Draft a re-engagement campaign and put a plan card on the board with the angle you'd take and why.
Segment your audience with tags and groups
Ask Claude to translate the brief into Mailchimp tags and groups and create them. A good starting set: engaged subscribers, lapsed (no open in 90+ days), new signups, and past buyers. Each one becomes an Audience card on the board, so you can see the logic and the size before anything gets built on top of it.
Create these Mailchimp segments and add an Audience card for each: engaged (opened in 30d), lapsed 90d, new signups (last 14d), past buyers. For each, note the hook and the offer you'd use, and what we should never say to them.
Draft the campaign or Customer Journey
Now the writing. Have Claude draft a one-off campaign, a recurring newsletter, or a full Customer Journey. Each email arrives as a card with subject-line variants, preview text, body, and CTA, linked back to the segment it targets. Because the lineage is visible, you can tell at a glance whether the offer fits the audience.
Draft a 3-step welcome Customer Journey for new signups. Email 1 welcomes and sets expectations, email 2 shares the best content, email 3 makes the first offer. Two subject lines each, warm not pushy. Put each step as an Email card linked to the journey.
Approve on the board: the safety layer
This is the step that makes an autonomous agent safe to run on a real list. Every email sits on the canvas as a finished, sendable asset with its full lineage: the audience, the goal, the offer. Read it, tweak the copy in plain language, then approve. Only approved cards get scheduled in Mailchimp.
The agent can draft a month of emails in a minute. The board is where one human decides which of them are good enough to send.
Pull results back and let the next campaign compound
After a send, ask Claude to pull Mailchimp opens, clicks, and conversions back onto the board as Result cards. Now the next campaign doesn't start from a blank chat. It starts from "the short subject line beat the clever one, and Tuesday morning won, do that again." Your learnings stop evaporating.
The campaigns and journeys to build first
If you do nothing else, build these. They have the clearest purpose, the best data to draw on, and the fastest payback, which makes them the perfect first job for the agent.
- Welcome journey. A Customer Journey that fires on signup. Claude leads with what new subscribers actually came for, sets the cadence, and times the first offer to step three instead of email one.
- Regular newsletter. Your recurring campaign to engaged subscribers. Brief Claude on the issue and it drafts the whole thing, with two subject lines and preview text, ready to schedule.
- Promotional campaign. A one-off to a buyer tag. Claude matches the offer to the segment and writes the urgency in honestly, without the all-caps countdown clichés.
- Re-engagement. A journey for the lapsed tag (no opens in 90+ days). Claude reasons about the right incentive instead of defaulting to the biggest discount, and suggests when to sunset the unresponsive.
- Date-based / birthday automation. A journey triggered on a signup anniversary or birthday merge field. Low effort, high open rate, and easy for Claude to template once and reuse.
How this compares to Mailchimp's AI and a plain chatbot
Mailchimp's own AI is genuinely handy. So is asking ChatGPT for a subject line. Neither replaces a stack that reads your audience, reasons across the whole campaign, and keeps a human in the approval loop.
| Approach | What it's good at | Where it falls short |
|---|---|---|
| Plain chatbot copy/paste into Mailchimp | Fast first drafts of copy and ideas. | No audience context, no memory between chats, every campaign starts from zero. |
| Mailchimp built-in AI | Subject-line and content help inside the Mailchimp editor. | Doesn't plan or write the whole campaign across your audience and offer. |
| Claude + Mailchimp + Plainpaper | Plans and writes the whole campaign, keeps lineage, approve before send. | You still set the strategy and approve, exactly as the workflow intends. |
Pitfalls to avoid
- Letting the agent send unreviewed. Keep send and schedule actions behind board approval. An agent that drafts is an asset; an agent that ships unreviewed is a liability.
- Skipping the segment step. Blasting the whole list is what makes people unsubscribe. Let Claude build the tags first so each email goes to people it's actually for.
- Recycling the same newsletter. If every issue is last issue with the date changed, opens decay. Brief Claude on what's genuinely new each time.
- One giant prompt. Work stage by stage (segment, then campaign, then copy) so each artifact is reviewable instead of a wall of text.
Key takeaways
- Connect Claude to Mailchimp (read + write) and a Plainpaper board over MCP.
- Brief Claude like a marketer (the goal, the offer, the audience) before it writes a word.
- Segment first, then build the welcome journey, newsletter, and re-engagement path; they pay back fastest.
- Approve every email on the board. The agent drafts; you decide what ships.
- Pull results back so each campaign compounds on the last instead of starting from a blank chat.
Frequently asked questions
Can Claude build Mailchimp campaigns?
Yes. Connected over MCP, Claude can read your Mailchimp audience and draft full campaigns: subject lines, preview text, body copy, and segments. It produces finished, approvable drafts, and a human approves them on the board before anything is scheduled or sent from Mailchimp.
Can Claude write my newsletter?
Yes. Brief Claude on the issue (the theme, the links, the call to action) and it drafts the whole newsletter as an Email card, with a couple of subject-line options to choose from. You edit the copy in plain language on the board, then approve it for the send.
Does it work with Mailchimp Customer Journeys?
It does. Claude can plan and draft the emails for a Customer Journey (a welcome series, a re-engagement path, a date-based automation) and lay each step out as a card so you can see the whole sequence before it's built and turned on in Mailchimp.
Is there an official Mailchimp MCP server?
Not for the Marketing API. Mailchimp publishes one official MCP server and it covers Mailchimp Transactional (Mandrill), the service behind receipts and password resets. Campaigns, audiences, and journeys live in the Marketing API, and to reach those from Claude you use a community MCP server, a hosted connector such as Zapier, or the Marketing API called directly. Anything advertised as a full Mailchimp marketing MCP server is third party, so read what it does before you hand it a key.
Does this work in Claude Code?
Yes, and Claude Code has the wider choice of the two. You can add a
Mailchimp MCP server with claude mcp add, or skip MCP and let
Claude Code call the Marketing API itself, since it is a REST API with
basic auth. Both routes reach the same endpoints for creating the draft,
setting its content, and scheduling the send.
Will Claude send without my approval?
No. The workflow is built so the agent only ever drafts. Every campaign and journey email lands on the Plainpaper board as a finished asset with its lineage (the audience, the offer, the goal), and you approve it there. Only then does Claude schedule or push it through Mailchimp.
Is this better than Mailchimp's built-in AI?
They solve different problems. Mailchimp's AI helps inside the editor: subject-line suggestions, content blocks, send-time tips. Claude reasons across the whole campaign. It reads your audience, plans the segments and the sequence, and writes every email. Plainpaper keeps that work visible and approvable instead of trapped in a chat window. Use both.
Do I need to know how to write prompts?
Not really. You brief Claude the way you'd brief a marketer (the goal, the offer, the audience) and steer from there. The board shows you exactly what it produced, so you correct in plain language instead of re-engineering a prompt.
Next step: give the workflow a home
You have the workflow. The piece that makes it safe and repeatable is the board in the middle, where the agent's work becomes visible, linked, and approvable. Plainpaper is free to try and connects to your Claude client over MCP in a couple of minutes. Your next Mailchimp campaign can start from last campaign's numbers instead of a blank chat.
Keep reading
- Mailchimp MCP server guide: the three real routes into Mailchimp from Claude, and the API-key setup the workable one needs.
- Welcome series playbook: the first flow to build once Mailchimp is connected, as a board with the five beats already written.
- Ecommerce campaign playbook: the end-to-end board: brief, audience, strategy, emails and creative, then results against the goal.
- Add Canva to this stack: the same workflow with the design layer, so the email arrives with its own visuals.
- Draft email campaigns in Plainpaper: what an email card holds, how sequences and A/B variants work, and the handoff back to Mailchimp.
- Email tools that build templates from chat: the ESP connectors that can write a template for you, and why Mailchimp's is not one of them.
- Install Plainpaper in Claude: the connector setup, about two minutes, no install.