The Slovenian property register as a tool your assistant can call

Why a server rather than a ZIP download

• 3registers

  Three registers behind one sign-in

  The transaction register (sales and leases since 2007), the property cadastre (addresses, parcels, buildings, units) and the valuation register. Each publishes on its own schedule, in its own format, through its own interface. Here they are one mirror and one token.

• 1000rows per query

  Real SQL, with a guard in front of it

  The query_register tool runs one SELECT against the mirror and returns at most 1,000 rows. Everything else is refused before it reaches the database. The schema is published as a register://schema resource, so the model is not guessing column names.

• 60calls a minute

  The same answer our own assistant gives

  The server exposes the tools our own chat uses, so the answer is the same wherever you ask it. A call log is kept for 90 days; the contents of a query are not kept at all.

• 0lines of code

  OAuth, or an API key in one header

  In Claude and ChatGPT you add the server by address and sign in over OAuth, with no key to create. In Claude Code, Cursor and VS Code you paste an API key into a single header. Keys are made and revoked in your account.

What this is

The Model Context Protocol (MCP) is an open way to give an AI tool access to data it does not carry itself. Our MCP server opens the Slovenian property transaction register (prices actually paid for sales and leases since 2007), the property cadastre (addresses, parcels, buildings, units) and the public valuation register to your own assistant.

The server address is:

https://dragaslovenija.si/mcp

The data is public and published under CC BY 4.0 by the Surveying and Mapping Authority. What a subscription pays for is the mirror being kept current, the guard over the SQL, and a server you do not have to run.

Thirteen tools

They are the same tools our own assistant uses, so the answer is the same wherever you ask it.

• search_place find a municipality or settlement and get the IDs the register uses
• lookup_address one street address: the building, its parcels, its units and the generalised market value from the valuation register
• lookup_parcel one parcel: area, designated and actual land use, land quality rating, protection regimes, and the owner where the owner is a company
• find_parcels a list of parcels read live from the cadastre: buildable, empty, over a given size, by municipality, cadastral municipality or radius
• lookup_building one building, by cadastral municipality and building number
• nearby_deals real transactions within a radius of a point, with the median, the quartiles and a breakdown by year
• query_register one read-only SELECT over the register mirror, at most 1,000 rows
• market_stats medians per year for a municipality and category, sales and rent, with the gross yield
• construction_activity reported construction starts in a municipality or within a radius, new builds and removals counted separately, month by month
• spatial_acts the spatial acts (OPPN) over one point, parcel, address or municipality, with the zoning from the OPN
• find_projects our directory of new-build developments: what is being built and sold, with the developer, the asking price and the expected completion
• neighborhood the surroundings of one point: flood hazard, landslides, erosion, seismic ground acceleration, the age of the buildings nearby and what stands around it in OpenStreetMap
• map_link a link into our map, filtered or on one transaction

Alongside the tools the server offers a register://schema resource carrying the full schema and the query rules, and a ready-made analyse_address prompt that works through one address in order: the cadastre, the modelled value, real nearby transactions, and the municipal comparison.

What a session looks like

Three questions, the tool call each one becomes, and the shape of what comes back. The figures below are illustrative; the tool names, their arguments and the fields are the server's own.

market_stats { "municipality": "MARIBOR", "category": "apt", "from": 2021 }

{ years: [
    { year: 2021, median_eur_m2: 1620, n_priced: 1104 },
    …
    { year: 2025, median_eur_m2: 2180, n_priced: 1287 } ],
  rent: { median_eur_m2_month: 8.1, year: 2025 },
  gross_yield_pct: 4.46 }

lookup_parcel { "cadastral_municipality": "RUDNIK", "parcel_no": "2389/5" }

{ area_m2: 612, planned_use: "residential",
  actual_use: "built-up land", soil_score: 0,
  regimes: ["water protection zone III"],
  value_eur: 214300, owner_type: "company" }

nearby_deals { "lat": 46.0512, "lon": 14.5054, "radius_m": 300, "category": "apt" }

{ n: 34, median_eur_m2: 5240, p25: 4610, p75: 5980,
  by_year: [ { year: 2024, n: 11 }, { year: 2025, n: 14 }, … ],
  deals: [ { signed: 20250918, area: 62.4, eur_m2: 5310, distance_m: 84 }, … ] }

How to connect

Claude (claude.ai and the desktop app)

In settings choose Connectors, then Add custom connector, and paste the server address. Our sign-in and consent page opens; there is no key to create, because the sign-in runs over OAuth.

https://dragaslovenija.si/mcp

Claude Code

With an API key made in your account:

claude mcp add --transport http dragaslovenija https://dragaslovenija.si/mcp \
  --header "Authorization: Bearer ds_live_..."

Or with the address alone; /mcp inside a session then runs the OAuth sign-in:

claude mcp add --transport http dragaslovenija https://dragaslovenija.si/mcp

Cursor

In ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "dragaslovenija": {
      "url": "https://dragaslovenija.si/mcp",
      "headers": { "Authorization": "Bearer ds_live_..." }
    }
  }
}

VS Code and GitHub Copilot

In .vscode/mcp.json:

{
  "servers": {
    "dragaslovenija": {
      "type": "http",
      "url": "https://dragaslovenija.si/mcp",
      "headers": { "Authorization": "Bearer ${input:ds_key}" }
    }
  }
}

ChatGPT

Add the server by address under Connectors in settings. ChatGPT signs in over OAuth, the same way Claude does; its documentation does not describe a custom Authorization header, so an API key is not the route there.

API keys are created, listed and revoked in your account, under MCP and API. A key is shown once, when it is made. If you lose it, make a new one and revoke the old.

Limits

• 2,000 calls a day per account, and at most 60 a minute per key.
• One SQL query returns at most 1,000 rows and may only SELECT. Everything else is refused before it reaches the database.
• The cadastre tools read the public GURS services live, so they are slower than a query against our own mirror and depend on those services being up.
• The server cannot read your alerts, chats, reports or listings, and it cannot change anything.
• A call log (when, which tool, how long, whether it worked) is kept for 90 days. The contents of a query are not kept.

Who it is for

• Analysts and researchers

  SQL over the whole property transaction register, with no ZIP file to download and unpack every time a new edition lands. "Which streets in Kranj hold the most houses built in the seventies?" is one query rather than one evening.

• Valuers

  An address resolved, the comparable transactions in a radius, and the generalised market value, without switching between three public viewers. What comes back is what the register shows at that moment, and the source is named with it.

• Investors and developers

  A parcel, its designated and actual land use, its land quality rating and its protection regimes, alongside real nearby transactions and the municipality's price trend. All inside a conversation you can keep asking questions in.

What it costs

MCP access comes with the Pro plan, alongside the market reports and the listing creator. The data itself is public and free; what a subscription pays for is the mirror being kept current and a server you do not have to run.

Cancel with one click in your account. API keys can be revoked one at a time.

Prices include VAT. For businesses with a valid EU VAT number from another member state the reverse charge applies: the buyer accounts for the tax.

Frequently asked questions