The short version
Before MCP, getting portfolio data into an AI assistant meant exporting a spreadsheet and uploading it. The assistant saw a frozen copy with no context and no permissions.
With MCP, the assistant asks the application directly. The application checks who you are, runs its own calculation and returns the answer with links back to the source. Nothing is copied into the chat that you could not already see.
How it works
- 01 /The application runs an MCP server at a web address, for example your QTR connector address.
- 02 /You add that address to your assistant as a connector.
- 03 /The assistant sends you to the application’s sign-in page. You approve what it may do.
- 04 /During a chat, the assistant calls the application’s tools, such as “budget variance” or “create action item”, when your question needs them.
- 05 /The application answers under your permissions and records any change it makes.
Tools, reads and writes
An MCP server publishes a list of tools. Read tools return information. Write tools change something. In QTR, more than 170 tools cover financials, the rent roll, leasing, collections, comps, work orders, Action Center, reports and deals.
What it means for permissions
- You sign in as yourself, so the assistant has your access and no more
- Read and write are separate grants; write is off unless you choose it
- Your company admin can turn connector access off for any user
- Removing access stops the connector on its next call
Which assistants support MCP
Claude, ChatGPT, Grok and Cursor all connect to QTR today. Support for remote MCP servers is spreading across assistants, so the same connector works with new clients as they add it.
