Your data has answers.
Empower anyone to get them.
Providing conversational access to your data through MCP servers
Conversational access to your data — MCP servers
The problem
The data is there.
Access is the bottleneck.
Most organizations sit on far more information than they can practically use.
A planner needs one figure from a geodata layer and has to open a GIS tool to get it. A transport team wants to cross-check schedules against fleet data held in two different systems. An operations manager knows the answer exists somewhere in the asset database but can’t extract it without help from IT.
The result is the same everywhere: rich data, locked behind specialist tools and specialist skills. The people who need the answers can’t reach them, and the people who can reach them spend their time running other people’s queries.
What is an MCP Server?
A safe bridge between an AI assistant and your data
MCP — the Model Context Protocol — is an open standard that lets an AI assistant connect directly to a live data source.
Without it, an AI assistant can only guess from what it was trained on. With it, the assistant queries your actual data and answers from what’s really there: current, accurate, and traceable back to the source.
An MCP server is the connector that makes this possible. It sits between the assistant and your database, translating a plain-language question into a precise query and handing back an answer based on your data. You keep full control of what the assistant can see and do.
A concrete example: the Canton’s SITG Geodata
We built one for Geneva’s territorial data
To show what this looks like in practice, take a data source every organization in the canton already knows: the SITG — the Système d’Information du Territoire à Genève.
For 25 years, a network of 16 public partners has collected and shared Geneva’s geographic data through the SITG: cadastre, zoning, buildings, energy, natural hazards, pollution, trees, and hundreds of other layers, all freely available. It’s an extraordinary public resource, with over 1000 rich data sources, it rewards those who know exactly how to navigate it.
Before, 6 steps
- Open the SITG map viewer,
- Identify the correct layer among thousands,
- Apply the right filter,
- Combine the datasets in a GIS application,
- Run analysis,
- Repeat for the next question.
After, 1 step
Natural language questions:
- “What’s the zoning for this address?”
- “What is the solar potential for buildings consuming a lot of energy?”
- “Show me the buildings in this sector connected to the heating network.”
The assistant returns the answer along with a map, tells you which SITG layer it came from and gives you back the script it ran — so it’s verifiable, current, and grounded in official cantonal data. No GIS software, no query language, no specialist required.
The data happened to be geospatial. The approach doesn’t care.
Where else this plugs in
Same pattern, different data
The SITG example is geographic because that’s where Novel-T’s roots are.
But an MCP server can sit on top of almost any structured data source. The interface stays the same — you ask, it answers — while the data behind it changes.
Another geographic database
Point it at a different canton’s geodata, a utility’s network maps, or a company’s internal cadastral and asset-location data. The same conversational access, on your territory.
A public transport entity
Schedules, fleet status, ridership figures, infrastructure and asset registries — often spread across separate systems. One assistant that can answer across all of them, for operations teams and the public alike.
Operational & asset-management systems
Maintenance records, equipment registries, inventory, sensor data. Let field and management staff ask direct questions instead of waiting on a report.
See it on your own data
The fastest way to understand what this can do is to see it running on data you recognize.
Book a short demo and we’ll show you the SITG assistant in action — then talk through what the same approach could look like on yours.
See more info here.
