io.github.7nashinick/norwegian-data-mcp

norwegian-data-mcp

Norwegian public data in your AI assistant: companies, statistics, addresses, property. One MCP server, zero API keys, zero signup.

Ask Claude (or any MCP client) things like "who is on the board of Equinor and what was their revenue last year?" or "how has Norwegian inflation moved this year?" and it answers from the official registers, with sources.

Install

Claude Code:

claude mcp add norway -- uvx norwegian-data-mcp

Claude Desktop (claude_desktop_config.json):

{
  "mcpServers": {
    "norway": {
      "command": "uvx",
      "args": ["norwegian-data-mcp"]
    }
  }
}

Any other MCP client: run uvx norwegian-data-mcp (or pip install norwegian-data-mcp and run norwegian-data-mcp) as a stdio server.

Tools

Try these

• "Who is on the board of Equinor, and what was their revenue last year?"
• "Find all breweries in Bergen. Any of them bankrupt?"
• "What's Norway's inflation rate right now, and how does the last year look?"
• "Population of Oslo over the last 10 years."
• "What are the coordinates and gnr/bnr of Karl Johans gate 22, Oslo?"
• "Norwegian mortgage and policy interest rates over the last two years."
• "How has the krone moved against the dollar and euro this month?"
• "Map Equinor's subsidiaries with ownership stakes."

How responses look

Every tool returns a compact JSON envelope: the data (statistics come as CSV, the most token-efficient table format), a hints list that tells the model how to narrow or continue, and a source string naming the register and license. Responses are capped in size; SSB's dense JSON-stat2 cubes are flattened to plain tables with time on the rows.

When SSB closes a table (it happens: CPI table 03013 was replaced by 14700 in 2026), the server reads the succession note and follows or suggests the successor instead of failing.

Data sources and licenses

All four are open government data with no key or registration. Every response carries its attribution, so the model can cite it.

What this server will not do

• Property owners, buildings, valuations, grunnbok: Norway keeps these behind a licensed Matrikkel agreement that private individuals cannot apply for. No open API exists, so no tool pretends otherwise.
• Shareholder registers and tax lists: no open APIs exist; tax-list searches are logged and the searched person is notified.
• Person search: looking up all roles a person holds across companies is legally restricted (enhetsregisterloven §9-1). Per-company roles are the lawful slice and that is what get_company_roles serves.
• Historical financials: the open Regnskapsregisteret API serves the latest filed year only; the tool links the official annual-report PDF for the rest.

Development

python -m venv .venv && .venv/bin/pip install -e '.[dev]'
.venv/bin/pytest                                # offline, fixtures only
.venv/bin/pytest -m live --override-ini addopts=''  # live smoke against the real APIs

Runtime dependencies are mcp and httpx, nothing else. Responses are cached with per-source TTLs (in memory plus ~/.cache/norwegian-data-mcp, disable with NORWEGIAN_DATA_MCP_CACHE=off), and a stale cached copy is served with a note if a register is briefly unreachable.

Roadmap

• plot area via the Teig WFS, daily policy-rate series, elevation and kommune info tools
• streamable HTTP transport for hosted use