How to evaluate an email creation platform's MCP server

In 2026, many teams are looking to consolidate more of their work directly within AI apps like Claude, ChatGPT, Gemini and Microsoft Copilot. If this is you, then you are probably trying to make sure every new vendor you onboard has an MCP server.

Many email creation platforms have released an MCP server, or are in the process of building one. But not all MCP servers are created equal, and this article covers what actually separates a serious one from a feature shipped to tick a box.

Look past the checkbox

Avoid treating “has an MCP server” as a box to tick off when comparing vendors. Requiring one is reasonable, but it doesn’t go nearly deep enough. MCP is a fundamentally new category of software, and if you’re serious about adopting AI, it is just as important as the UI going forward. You’d never put “has a UI” on a checklist, because of course a platform has one. What matters is what it can actually do, and how productive it makes your team. That is how you should think about an MCP server: it’s the difference between a fun gimmick and something that can seriously transform your team’s workflow.

So, what can it actually do?

If you are evaluating an email creation platform vendor, and are moving more of your team’s work to apps like Claude, or are planning to over the next year, there’s not a whole lot of point in evaluating the wider platform’s capabilities in isolation: what really matters is what you can actually do with the MCP server.

Building + editing emails

At a minimum, you’ll want it to be able to build emails for you via chat. And those emails should strictly adhere to your brand, your guardrails, your approved component library (your Email Design System, or other brand controls). This is non-negotiable: if it doesn’t strictly adhere to your brand, then you will spend just as long fixing its output to ensure that it is consistent, and that it renders correctly in different clients.

Make sure you can also edit existing emails, not just generate new ones: some vendors let you generate an email, but any further changes then have to be made manually within the platform rather than by sending further prompts.

It’s also worth checking what it can’t do. Is there anything you can do in the platform’s editor that can’t be done via the MCP server? Even small gaps in building capability versus the platform’s UI can slow you down, since you’ll be jumping back into the platform to make tweaks by hand. Can it apply dynamic content to your emails? Can it search and use images from your approved image library? Can it localize and translate emails?

Something worth confirming is if the emails built using the MCP are fully editable in the platform’s email editor — you never know when you might need to make manual tweaks, without AI.

The wider workflow

If you’re serious about email and reinventing your workflow, you’ll know there’s far more to email than the build. Have a think about which other parts of your workflow you’d benefit from being able to do via chat, and ask your vendor if their MCP server supports them. Below are just some examples:

• Build and edit your Email Design System components, design tokens, brand controls and other guardrails?
• Create and manage dynamic content and your image library?
• Send test emails?
• Collaborate with colleagues — comment, request and submit feedback, publish?
• Export straight into your ESP/MAP?
• Download images and PDFs?
• Search and organize your emails across projects and workspaces?

It might be helpful to literally write down the prompts that you’d like to be able to put to your AI assistant, and ask your vendor if they would work. Here are some examples:

“Build out an email based on my brief in Asana, and send me a test email once you’re done.”

“Can you add a new 3-column card block to my design system? Make it similar to the existing 2-column card block, and make sure the headline, image, image alt text and body copy are all editable.”

“Can you read through all the feedback/comments I’ve received on this email and address each point?”

“Can you review this email against our brand guidelines and add some comments if there are any issues? If not then approve it.”

“Can you export all the emails in this project to Braze?”

“Can you download a PDF of every email in this project and drop them in my Google Drive for legal to sign off?”

Most vendors’ MCPs today only focus on the email building aspect, but as you can see there are lots of other parts of the workflow to consider.

Does it keep up with the platform?

One of the questions worth asking the vendor you are evaluating is: what is their attitude to adding capability to the MCP server?

Some vendors build an MCP server to tick the box, ship it with a handful of features, and move on to the next thing.

What you want to know is: when they develop new/future platform features, do they add them to the UI and the MCP automatically? Or is it just the UI - and anything being added to the MCP in future is an afterthought? It’s useful to get a feel for this, as you might end up having to chase them down to add things.

Can you render a preview directly within the AI app interface?

Building emails is ultimately a visual task. You will want to see what the email looks like, and indeed how it changes as you send further prompts, so it’s important that the MCP server can render a preview from directly within the AI assistant’s interface. Some vendors do not let you see a preview directly in the interface; they just put a link to the platform. This can seriously kill the flow if you’re having to check the preview in a platform between each prompt.

Remote vs local

There are two types of MCP servers: remote and local. There are legitimate uses for local MCP servers, but for an email creation platform, generally the one you want is remote. A remote MCP server runs in the cloud, meaning there’s nothing to install and no software to configure on each individual user’s machine.

A local MCP server is more typical of developer tooling. It runs on your machines directly. As such it needs software packages installed and kept up to date on each person’s machine. If you’re an enterprise team of thirty, that’s thirty installs to set up and maintain, and your security team will be far from thrilled about it. In practice, some security teams respond by locking the MCP down to developers, which means the rest of your team never gets to use it at all. If the vendor you are considering only has a local MCP server, it’s worth checking with your security team beforehand what they will actually allow.

Governance, permissions and auditability

How are permissions handled by the MCP server? The gold standard is OAuth 2.1; under this strategy, your users will log in once and then the permissions they get are inherited from the in-platform permissions, ensuring nobody can access areas that they wouldn’t be able to within the app.

As a bonus, admin controls to restrict this further are a nice to have: as a rule you may want to configure the MCP integration not to be able to delete projects or emails, for example, or to restrict it to read-only access.

Another question worth asking is: is there a way to audit what has been done via AI?

You should be suspicious of having to configure things like API keys for an MCP server: an MCP server requiring a manual API key is a good indication that it does not use OAuth, and that means everyone using that API key would be on the same permissions, with the same access, and no way to audit who has done what.

Watch out for hidden costs

Watch out for hidden costs: the way some vendors build their MCP is that it hands the task off to its own AI internally. The problem with this is that you could end up paying double: for the tokens from your own AI provider, but also for credits from your email platform. Check with them if there is any cost to using the MCP server.

How EmailShepherd approaches MCP

At EmailShepherd, every feature that is available in the UI is also available via MCP. We maintain full parity. MCP is treated as a first class citizen alongside the UI: that means that for every new feature we build, our process is to consider both how to add it to the UI, and how to add it to the MCP.

Questions to ask

Take these into your next demo.

Area	Ask this	What a strong answer tells you
Brand & guardrails	Does it build strictly within my Email Design System, approved components and brand controls?	It only ever produces on-brand, client-safe emails, so you’re not fixing its output by hand.
Building & editing	Can it both create new emails and edit existing ones from chat?	Both: you can create emails, or edit existing ones and keep refining via prompts.
Building & editing	When building or editing emails via chat using the MCP server, can you do everything you can do in the editor?	Full parity with the UI editor: anything you can do in the editor you can also do via chat.
Building & editing	Are emails built via the MCP still fully editable in the platform’s editor?	Yes: you can always drop into the editor and make manual tweaks, without AI.
Wider workflow	Beyond building, can you do the rest of your workflow from chat too — design system changes, test sends, approvals, export, PDFs, etc?	Full parity: everything you can do in the UI, you can do from the MCP, not just the build.
Future features	What’s your process for adding new features to the MCP server?	Every new feature ships to the MCP at the same time as the UI, never as an afterthought.
Preview	Can it render emails directly in the AI interface?	You can see the email and watch it change as you prompt, without jumping back to the platform.
Deployment	Is it a local or remote MCP server?	Remote, so there’s nothing to install and your whole team can use it right away.
Permissions	How are permissions handled? Can I restrict what the MCP can do?	OAuth 2.1, so each user inherits their in-platform permissions, with admin controls to restrict further (e.g. no deleting projects or emails).
Auditability	Is there a way to audit what’s been done via AI?	A full, reviewable record of every action taken through the MCP.
Pricing	I already pay my own LLM provider to use it. Do I also have to pay for AI credits from you?	No double-billing; you pay for your own tokens and nothing extra on top.