Strata company: AI that answers questions from on-premise strata systems
A custom MCP server that connects a strata manager's on-premise MRI Strata Master and GetReal systems to Claude and Microsoft Copilot, plus a month-end invoice workflow that took a 40-hour job down to 4 minutes.
The problem: the answers were split across two on-premise systems
The business runs on two systems hosted on its own servers. MRI Strata Master holds the strata plans, lots, owners, contacts and insurance. GetReal holds the jobs, management activity and the tradespeople who do the work.
Neither had a ready-made connection to Claude or Copilot, and because both are on premise, an off-the-shelf cloud connector wasn't an option. Answering an everyday question, like which tradesperson is on a job, who owns a lot or what a building's policy covers, meant searching both systems and piecing the answer together by hand.
Month end was the heaviest load. Monthly management invoices arrive as PDFs, and reading and entering them took around 40 hours a month.
The solution: one connection to both systems, usable from Claude and Copilot
G3 Digital built a custom integration server that runs alongside the client's systems and connects them to the AI tools the team already uses: Claude and Microsoft Copilot.
1. An MCP server for Claude and Copilot
Four read-only search tools, for jobs, lot contacts, tradespeople and insurance, are exposed through the Model Context Protocol (MCP). Claude and Microsoft Copilot Studio both add the server directly as a tool, with no extra connector to maintain.
2. Tradesperson records kept in sync
Tradesperson details flow from GetReal into Strata Master. Clear matches update automatically, while conflicts and new contractors go to a review screen where a staff member approves, rejects or amends them.
3. Month-end invoices entered automatically
Monthly management invoice PDFs are read, checked so the totals add up, and entered into Strata Master. Anything that doesn't add up is flagged for a person instead of being entered.
The data stays on the client's own servers. The AI can only search through the integration, every request needs an access token, and changes to records go through the sync rules and the review screen.
What the team can ask
Because the AI searches both systems at once, one question can follow the links between a building, its jobs, the tradespeople and its insurance.
One question, both systems
"Has anything been done about the storm damage at this building? Which tradesperson is on it, has the quote gone to council, and what's the water damage excess on the building's policy?"
The answer pulls the job and its activity from GetReal, the tradesperson's details, and the policy and excesses from Strata Master.
"My swimming pool equipment is broken, is anyone dealing with that?"
Finds the pool pump job even though the question shares no words with it.
"Who owns lot 5, and is there an emergency contact?"
Returns the owners, their numbers and the extra contact on file.
"When does the strata policy renew, and is flood covered?"
Reads the renewal date, premium and policy notes.
"Is there a job for a leaking roof at a plan we don't manage?"
Says there's nothing on file instead of guessing.
Insurance claims are being added next, at the client's request.
The business return
The team kept the systems they already depend on. What changed is how quickly they can get answers out of them, and how little of month end is spent typing.