Capability
MCP interface for AI tools
Your own AI assistants work directly with Caymland M4. Through an open standard, with your permissions and approval before every change.
Your AI, your data, your rules.
An open standard.
No special project, no vendor lock-in.
Closely held.
OAuth, your permissions, mandatory approval, no bulk extraction of contacts.
It understands itself.
The assistant reads your structures instead of depending on a hard-built special case.
Your AI tools should be allowed to work with your data
An AI assistant is now in use in many companies. It can phrase, summarise, research. There is only one thing it cannot reach: your own figures.
So people copy. A report is exported, pasted into a chat window, evaluated, and the result carried back. In the process data leaves systems that were set up under control, and afterwards nobody knows exactly what went where.
Caymland M4 solves this in the right place: with an interface AI assistants understand.
Related capabilities with pages of their own: the AI assistant, which operates the system in conversation, and AI & machine learning for the patterns in your data.
An open standard, not a special route
Caymland M4 speaks Model Context Protocol (MCP), the standard through which AI assistants communicate with business systems.
In practice that means: you enter the connection in the AI tool your company already works with. From then on it can use your marketing data. No project of its own, no bespoke solution, no vendor you are tied to.
And because it is a standard, the same holds for the tool you use in two years.
What your assistant can do with it
| Area | What is possible |
|---|---|
| Search | Resolve a contact, company, campaign, email, segment, form, template, tag or account owner by name |
| List | "Which segments, campaigns and emails do I actually have?" |
| Analyse | Evaluate contacts, campaigns and emails, and transactions and bookings from connected modules too |
| Create | Create segments, campaigns, emails, contacts and tags |
| Refine | Adjust existing segments, campaigns and emails |
| Check | Retrace the change history and origin |
| Send | Test sends and real sends, expressly approved |
| Understand | Have the structure, fields and campaign build explained, so the assistant works on solid ground rather than guessing |
Practically that means: management asks in the AI tool it knows how recent campaigns have developed and gets the answer out of Caymland M4, without signing in and without anyone building a report.
What deliberately does not work
The limits matter as much as the possibilities. They were set on purpose, not forgotten.
No bulk extraction of contact data. People can be resolved deliberately when somebody asks about a particular customer. There is no route to retrieving the contact base as a whole. That removes the scenario every data protection officer rightly warns about.
No silent change. No connected assistant can create, change or send anything without a person having agreed.
No special rights. An assistant can do exactly what the person behind it may do. No more.
How it is secured
- Sign-in through OAuth. Access is granted in a regulated way and can be withdrawn at any time. No password sitting in a configuration file being passed around.
- Permissions apply unchanged. The rights of the signed-in person determine what is possible.
- Two-stage changes. The assistant proposes, a person approves. The approval covers exactly that one operation and cannot be redeemed a second time, not even on simultaneous requests.
- Protection of personal data. Before anything is handed to a language model, personal details are replaced and put back only in the answer.
- A complete log. Who asked what and when, and who approved it.
Why this is more than another API
A classic API addresses programs. Somebody has to read it, understand it and build something against it. That costs a project, and afterwards it can do exactly the one thing that was built.
An MCP interface addresses an assistant that understands it by itself. It sees what is possible, how your data is structured and which fields exist at your organisation. Then it combines that into whatever is being asked right now.
That is why you do not have to know in advance which questions somebody will ask.
What this looks like day to day
Management. Asks in the company's own AI tool: "How did our campaigns go last quarter, and where are we falling off?" The answer rests on the real figures from Caymland M4.
Sales. Asks about a customer's history before a meeting and gets a summary, without clicking through tabs.
The agency. Works with an assistant of its own, creates segments and drafts, and your team approves. The collaboration runs through controlled access rather than exports by email.
Your own automation. Your company builds an agent of its own for recurring evaluations. It uses the same interface, under the same rules.
Questions & answers
Frequently asked questions
What is MCP, actually?
An open standard through which AI assistants can talk to business systems. Similar to the way programs communicate through interfaces, but designed for assistants that work out for themselves what is possible.
Do we need this to use Caymland M4?
No. It is an additional route in. If your company does not work with AI tools yet, you lose nothing.
Which AI tools work with it?
Any that support MCP. By now that means the common assistants, as well as your own applications built on the standard.
Could somebody extract our customer data through it?
No. A bulk query of the contact base is deliberately not provided for, access is tied to a person and their permissions, and it can be withdrawn at any time.
Could an assistant accidentally send an email to everyone?
No. Every writing and sending action needs express approval by a person, and that approval is valid exactly once.
Do we see what happened?
Yes. Every request and every approval is logged.
Can we withdraw access again?
At any time, like any other access.
The difference in one sentence
