Artificial intelligence

Kinds now speaks MCP: connect your AI assistant to your security training

Connect your AI assistant to your security awareness training. Get answers, build your own dashboard or assign training, with a preview before every action.

The problem: one more console

Security awareness training is one more console nobody has time for. You open it to check who is overdue, to pull click rates for a report, to chase a few reminders. Then you close it and go back to the tools you actually live in.

Many admins already work in an AI assistant and ask it questions all day. So we stopped asking them to come to a dashboard and made Kinds reachable from the assistant itself.

What MCP is

The Model Context Protocol (MCP) is an open standard. It lets an AI assistant use an outside product through a set of named tools. The Kinds MCP server gives your assistant those tools, so it can read your Kinds data and take the same actions an admin takes in the app. Our glossary has a longer answer to what Model Context Protocol is.

What you can do

Four things, each with a real request.

Pull an answer

“Which departments had the most phishing clicks last quarter?”

The assistant asks Kinds, gets the numbers and replies with a table. If you report on phishing simulation results, this is the part you will use most. A few other requests that work the same way:

  • “Show click rate and report rate for the last six months.”
  • “List everyone who clicked a simulation in October, with their manager.”
  • “Who uses Salesforce and has not finished phishing training?”
  • “Are we covered for the frameworks we follow? Who is behind?”
  • “Is our directory sync healthy? When did it last run?”

Build a dashboard

“Make me a page showing training completion and phishing results by month.”

Your assistant builds the page from live Kinds data. Kinds does not design it for you. That is the point: it takes the shape you report upward in, not the shape we picked.

Export from a terminal

An engineer at an MSP asks a coding tool for last month’s overdue learners across every client organization, saved as a spreadsheet.

“How many of our people are overdue on training, by organization?”

The file lands where the engineer is already working. Nobody opens a dashboard.

Take an action

“Assign the phishing basics workshop to everyone who clicked last month.”

The assistant shows exactly who will be assigned and what will be sent. You confirm in the assistant, and Kinds does it. Reminders, the manager digest and removing people who left work the same way.

How connecting works

There is one address for every customer: https://mcp.kindssecurity.com/mcp. You add it in your assistant, sign in the way you already sign in to Kinds (Google, Microsoft or email), and approve the connection on a Kinds screen that names the assistant. This is OAuth, so there are no keys to create, paste or leak. Access passes are short-lived and renew quietly.

When you connect, you choose how much the assistant may do:

  • Read. The assistant can look but not change anything.
  • Read, create and delete. The assistant can also create, change and delete.

An admin who only has read access in Kinds can only choose Read.

Any MCP client can connect, including an app you build yourself. The Kinds MCP server has been tested with ChatGPT and Claude Code, and is designed for Claude (web, desktop and mobile), Cursor and VS Code as well, so you can connect Claude to your security awareness training the same way you would connect any other tool.

How actions stay under your control

Reading is simple. Acting is where you want care. Every action runs the same way:

  1. The assistant asks Kinds for the action.
  2. Kinds works out exactly what will happen (who it reaches, what gets sent, whether it can be undone) and returns that preview with a one-time code.
  3. You confirm where you are working: in the assistant, or in your own app.
  4. Kinds checks that nothing changed since the preview, runs the action once and records it.
  5. The assistant tells you what was done in plain words, and Kinds keeps the same line in its audit trail.

You never have to leave your assistant to finish something you started there. You confirm before anything runs.

Around that sit a few firm limits:

  • The same wall as the app. Every request goes through the permission checks Kinds already uses. A connection acts as the person who approved it and can never see or do more than that person can. One customer can never reach another’s data.
  • A record of everything. Each action is written to an audit trail with the person, the assistant and what was done.
  • Limits. There are hourly limits per person on how many actions and deletions can run.
  • Undo for deletions. Deleted learners can be restored for 30 days.
  • Three ways to stop it. An admin can disconnect any assistant, an account admin can switch MCP off for the whole account, and Kinds can switch it off for everyone.

What it cannot do

A feature list tells you what a product can do. This list matters more. Through the Kinds MCP server, an assistant cannot:

  • Manage admins or its own access.
  • Touch directory or Slack connections, or any password or key.
  • Edit workshops.
  • Change billing or delete the account.
  • Read what learners typed privately.
  • Change the audit trail.
  • Launch phishing simulations or edit phishing templates.

One more thing to be plain about. Your data does not stay inside Kinds when you use this. You choose the assistant, and names, email addresses and individual results go to that assistant, and so to the AI company behind it, under your own agreement with that company. Kinds does not send customer data to an AI company on its own initiative through this feature.

One set of tools, two doors

Clay is the assistant inside the Kinds app. The Kinds MCP server uses the same set of tools, more than 50 of them. Clay is the door inside the app; MCP is the door for the assistant you already work in. Anything one can do, the other can do. When we add a tool for one, the other gets it too. You can read more about what Clay can see.

For MSPs

If you run security awareness training for many clients, the console problem multiplies. One connection covers every client organization you manage, within the access you already have. Ask for last month’s overdue learners across every client in one request, or have your assistant produce a spreadsheet for a client review. If that is your setup, see how Kinds works for MSPs.

It is also a way to automate security awareness training with AI. Your assistant does the lookups and the paperwork, and you confirm the actions.

How to switch it on

Every account has the Kinds MCP server at no extra cost. It is off until an account admin switches it on in Settings.

  1. An account admin switches it on in Settings.
  2. Add the Kinds MCP server in your assistant, using the address https://mcp.kindssecurity.com/mcp.
  3. Sign in to Kinds as you normally do and approve the connection.
  4. Choose Read, or Read, create and delete.

Step-by-step instructions for each assistant are in the MCP setup guide.

Then ask it something.

Technical details

  • Open standard. Model Context Protocol. Both current versions of the standard are supported.
  • Sign-in. OAuth 2.1 authorization code flow with short-lived passes. There are no long-lived keys.
  • Scope. Read, or Read, create and delete, chosen per connection.
  • Permissions. Every call runs through the same permission checks, limits and audit trail as the in-app assistant.
  • Actions. Each action uses a preview and a one-time code tied to that exact preview.

Questions

Does it cost extra?

No. Every account has it.

Is it on by default?

No. It stays off until an account admin switches it on in Settings.

Which assistants work?

Any assistant or app that supports MCP, including one you build yourself. The Kinds MCP server has been tested with ChatGPT and Claude Code, and is designed for Claude, Cursor and VS Code as well.

Can it change things, or only read?

You choose when you connect: Read, or Read, create and delete. An admin who only has read access in Kinds can only choose Read.

Can I stop it?

Yes. You can disconnect the assistant, or an account admin can switch MCP off for the whole account.

Will the assistant see personal data?

Yes: names, email addresses and individual results, the same things you see as an admin. That data goes to the assistant you chose, and so to the company behind it.

Related reading