We connected an internal application to Claude through MCP, an open protocol that lets Claude talk to external systems. Users ask in plain words. Claude answers with the data.
Scenario
Your legacy systems already contain the data your team needs. The challenge is making that data available to AI without adding a chatbot or building a new interface into each application.
Opportunity
MCP lets us connect the system directly to Claude without changing the original application. Generative UI takes it a step further, letting Claude present each answer in the format that best fits the question: text, a table, a chart, or another visual.
We tested this with KZHoras, our internal system for tracking project hours and time off.
We built an MCP server that gives Claude access to its data, with permissions defining what each user can see. Users query KZHoras from a conversation, in Claude, without another AI interface.
Keeps its own interface, database and users.
Exposes only the data and functions we decide.
Where the question is asked and the answer is built.
The user asks Claude
They use natural language without needing to know how KZHoras is structured.
Claude calls KZHoras through the MCP server
The server receives the request and calls the KZHoras API.
Claude builds the answer
Claude queries KZHoras and answers in the format that fits the question: text, a table, a chart, or an interactive visualization you can click into.
A query, from the question in Claude to the answer built with KZHoras data.
Where the question is asked and the answer is built.
Exposes only the data and functions we decide.
The system of record for project hours and time off.
Runs the MCP Server.
No need to rebuild the application or create a new AI interface.
Users query data in natural language instead of learning reports, filters, or internal structures.
Claude becomes another way to access the existing system.
Results can come back as text, a table, a chart, or an interactive visualization.

