The Model Context Protocol is an open standard for connecting AI assistants to external systems. An MCP server exposes a product's capabilities as tools an assistant can call, which means Claude, ChatGPT or Cursor can read and act in that product rather than only talking about it.
For email marketing this closes a specific gap. An assistant with no access to your account can write you a subject line but cannot tell you whether last month was better than the month before, because it has never seen your numbers.
What changes when the assistant can see your account
The difference is not that the AI writes better email. It is that the questions you can ask stop being hypothetical.
Analysis without exports. "Compare September to August by campaign type, and tell me which segment drove the difference." Answered from your actual statistics rather than a CSV you assembled first.
List work without spreadsheets. "Add these forty leads to the webinar list, tag them webinar-october, and set their source field to the partner event." Done in a sentence.
Drafting with context. "Draft a campaign to customers who bought once in the last 90 days. Match the tone of the campaign that did best in August. Don't send it." The draft appears in your account, informed by what actually worked.
Diagnosis. "Which of my links got clicked most last quarter, and did any campaign have an unusual bounce rate?" The kind of question that is tedious enough through a dashboard that most people never ask it.
The value is mostly in the second category. Anyone can generate copy. The tedium of email marketing is list management, data entry and pulling numbers, and that is exactly what a tool-calling assistant is good at.
What it should never do unsupervised
An assistant with unrestricted access to an email platform is a liability, because the worst mistakes in email are irreversible. You cannot un-send a campaign.
A well-built MCP server draws the line in the tools themselves rather than in a policy document:
- Creating a campaign produces a draft. Nothing leaves the building without a separate, explicit send.
- Sending, deleting, unsubscribing and triggering automations require an explicit confirmation flag and are marked destructive in the tool metadata, so compliant clients ask you before running them.
- The server is stateless and holds no credential of yours. Your API token travels on each request, is used for that request, and is not stored.
That is how ours works. If you are evaluating any MCP server for a system that can send mail to thousands of people, those three properties are the ones to check.
Connecting one
There is nothing to install. A hosted MCP endpoint is a URL and a token.
For Claude Code:
claude mcp add --transport http igsendmail https://igsendmail.com/api/mcp --header "Authorization: Bearer YOUR_TOKEN"
For Claude Desktop, claude.ai and ChatGPT, add a custom connector pointing at the same URL with the same header. For Cursor, Windsurf and other clients, the equivalent block in your MCP config.
The token comes from your account settings, and it is scoped to your account, so the assistant can do exactly what you can do and nothing more. Revoke it any time.
Where this is going
The near-term effect is that the dashboard stops being the primary interface for the routine parts of the job. Pulling a report, adding contacts, checking a bounce rate and drafting a first version are all faster in conversation than in a UI, and none of them benefit from a screen full of charts.
The part that does not change is judgement. Deciding how often to send, what to claim, which segment deserves a different message and whether an email is good enough to represent you — none of that is delegated, and treating an assistant as though it were is the failure mode worth watching for. There is more on that boundary in AI email marketing.
The other thing worth saying plainly: an MCP server is only as useful as the data behind it. An assistant connected to an account with three campaigns and no custom fields will produce shallow answers, because there is nothing deeper to read. The value compounds with the quality of your segmentation and the fields you keep on each subscriber.
Ours exposes 32 tools covering lists, subscribers, campaigns, automations and the AI writing features, free on every plan including the free one, with the full list documented on the MCP page.
Frequently asked questions
What is MCP in email marketing?
The Model Context Protocol is an open standard that lets an AI assistant call tools in an external system. Applied to email marketing, it means an assistant can read your lists and campaign statistics and create drafts in your actual account rather than working from what you paste into the chat.
Is it safe to connect an AI assistant to my email platform?
It depends on the server. The properties that matter are that campaign creation produces drafts only, that irreversible actions require an explicit confirmation, and that the server does not store your credentials. Check those before connecting anything to a system that can send mail to your whole list.
Which AI assistants support MCP?
Claude Code, Claude Desktop, claude.ai, ChatGPT connectors, Cursor, Windsurf and VS Code in agent mode, among others. Support is spreading quickly because the protocol is open and not tied to any one vendor.
Do I need to install anything to use an MCP server?
Not for a hosted one. A hosted endpoint is a URL plus your API token in a header, added through one command or one configuration block. Servers that run locally need installing; hosted ones do not.
