Connect Claude to your business software with MCP

Asking an AI assistant about your own work used to mean copying data out of one app and pasting it into another. The Model Context Protocol (MCP) removes that step. It is an open standard that lets an assistant such as Claude call tools inside your software directly: look something up, read a record, and in some cases make a change you approve.

We build MCP into our own products from the start. This post explains how MCP connections work, what to check before you connect one, and how the three products we work on use it: Hark for meetings, Track My Test for labs, and Kelpic® for property management.

What an MCP connection actually does

An MCP server describes a set of tools to the assistant: their names, what they do and what they need. When you ask a question, the assistant decides which tools to call, the software runs them with your permissions, and the results come back into the conversation. The assistant never gets a copy of your database; it gets the answers to the specific calls it makes.

There are two common shapes:

• Local servers run on your own computer, next to a desktop app. Only assistants on that computer can reach them, and your data doesn’t need to be online at all.
• Remote servers run on the web, next to a cloud app. You add them to your assistant as a connector, sign in through the app’s own login page and approve what it may do. The connection uses OAuth, so you never paste a password or API key into the assistant.

What to check before connecting anything

• Read or write. Can the assistant only look things up, or can it change data? Writes are useful, but you should be asked to confirm them.
• Whose permissions. The assistant should act as you, with your access, and lose it when you lose yours.
• Scope. If you belong to several accounts or organizations, the connection should be limited to the one you choose.
• What leaves your systems. Tool results go into the conversation with your assistant, so they are covered by your AI provider’s data terms.

Hark: your meetings, on your computer

Hark records and transcribes meetings on your own computer. It runs a local MCP server while the app is open, so Claude Desktop or Claude Code (with your own Claude subscription) can read your meetings, transcripts and the live transcript of a meeting in progress, and start or stop recording. The MCP connection is part of Hark’s free features.

Typical questions: “What did we decide about the launch date in Tuesday’s meeting?”, “Draft a follow-up email from this morning’s call”, “List every action item assigned to me this week.” We go further in how to use Claude with your meeting transcripts.

Track My Test: lab specimens and freezers, read-only

Track My Test is our web app for labs: chain of custody for every specimen, freezer inventory down to the box position, and turnaround times. It has a remote MCP server. A lab member adds it to Claude as a custom connector, signs in with their Track My Test email, and picks which lab the assistant may read.

The connection is read-only on purpose. Lab history is a chain of custody, so every change should come from a person at a scanner or in the app. The assistant can answer questions such as “Where are study 12’s samples?”, “What’s expiring this week?” or “What was our median turnaround last month?“. It can’t move, ship or dispose of anything. Removing someone from the lab also cuts off their assistant, and there is no patient information to expose, because Track My Test doesn’t store any. How the AI connection works has the details.

Kelpic: property management, with changes you confirm

Kelpic®, property-management software from our sister company Kelpic LLC, also has a remote MCP server. When you connect it, you choose whether the assistant may only read or may also make changes. Reads cover things like rent bills, leases, maintenance requests and applications: “Who is behind on rent?” or “Which leases end in the next 90 days?“. A change, such as opening a maintenance request, is filed under your name, and your assistant asks you to confirm before it happens. Kelpic’s AI assistant setup guide walks through Claude, Claude Code and ChatGPT step by step.

The same idea in three places

	Hark	Track My Test	Kelpic®
Where the server runs	Your computer	trackmytest.com (web)	kelpic.com (web)
How you connect	Claude Desktop or Claude Code on the same computer	Custom connector, sign in with email code	Custom connector, sign in
What the assistant can do	Read meetings, control recording	Read only	Read, and changes you confirm
Limited to	Your computer’s meetings	One lab you choose	Your organization

The pattern is the same each time: the assistant works through the software’s own permissions, you can see what it is allowed to do, and the data stays in the system that owns it. That is the standard we hold our own software to, and a reasonable one to ask of any app you connect.