norapi-mcp
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Status: under active development. Everything described here works and is deployed, but the project is days old. Entry ids and the JSON shape may still change, and the curated list wants to grow. If you know a Norwegian API that is missing, suggest it.
Why
Norway's public sector publishes hundreds of APIs, but finding the right one means knowing which agency owns it, what it is called internally, whether it needs Maskinporten, and where the documentation moved to. Felles datakatalog holds the official metadata; it is written for registries, not for someone with a terminal open.
norapi takes that catalogue, adds hand-curated entries for the APIs people actually reach for, fetches the OpenAPI documents where they exist, and serves the result four ways.
Related MCP server: brreg-mcp-server
Four front doors
What you get | How | |
Web | One page per API with access, auth, formats, docs, a first call that works, and the endpoints from its OpenAPI document. Instant search and a ⌘K palette. | |
CLI | The same search locally, plus operation search across every indexed specification. |
|
MCP | Five tools for an AI assistant: |
|
llms.txt | One line per API, ready to paste into a prompt or a system message. |
Quick start
git clone https://github.com/r33hab/norapi && cd norapi
uv sync
uv run norapi search "kjøretøy" # full text, Norwegian or English
uv run norapi search "weather" --access open --auth none
uv run norapi ops "vegobjekter" # endpoints: method + path + summary
uv run norapi show brreg-enhetsregisteret # everything, including a quick_start callbrreg-enhetsregisteret
Enhetsregisteret API [Brønnøysundregistrene] (open; auth: none; JSON)
Oppslag og søk i alle norske enheter og underenheter: organisasjonsnummer, navn, …
https://data.brreg.no/enhetsregisteret/api/docs/index.htmlThe committed registry/ files are used as is. To refresh them from the sources:
uv run norapi harvest # Felles datakatalog + Geonorge access -> data/raw/fdk/ (a minute or two)
uv run norapi build # curated + harvested (+ cached specs) -> registry/
uv run norapi enrich # OpenAPI documents -> specs.json + operations.json
uv run norapi check --curated-only # probe docs, spec and endpoint URLs, record the verdicts
uv run norapi samples # run the front page's five example calls, keep the answers -> samples.json
uv run norapi site --out _site # the static siteA GitHub Actions workflow does the same every Monday and commits the result.
What is inside
Source | Entries | Notes |
Felles datakatalog (Digdir) | about 960 | Everything agencies have registered as a data service, with endpoints, specs, licences and keywords pulled over SPARQL |
Curated YAML in | 708 | Brønnøysundregistrene, SSB, Entur, MET, Kartverket, Nav, Skatteetaten, Digdir, Altinn, Vegvesen, Norges Bank, Stortinget, NVE, Udir, FHI, Helsedirektoratet, Lovdata, Elhub, Sikt, NDLA, Nasjonalmuseet, Norid, bank and payment APIs (SpareBank 1, Vipps MobilePay, BankID, open banking aggregators), Tolletaten, Fiskeridirektoratet, Finanstilsynet, Vinmonopolet, Luftfartstilsynet, county and municipal map servers and more, each with access, auth, formats, docs and a |
OpenAPI documents | 340 specs, 7 416 operations | Resolved behind |
Auth from an anonymous call | 747 harvested entries | Felles datakatalog has no auth field; |
Every curated entry is link-checked (ok, answers, missing) and the verdict is shown on its page.
How it is built
registry/curated/*.yaml ──┐ registry/registry.json ─┬─▶ norapi search / ops (CLI)
data.norge.no ─▶ harvest ─┼─▶ norapi build ─▶ registry/llms.txt ├─▶ norapi-mcp (AI agents)
OpenAPI docs ──▶ enrich ──┘ registry/operations.json└─▶ norapi site ─▶ _site/ (web)Harvest reads Felles datakatalog through its search API (the list) and SPARQL endpoint (the fields the list leaves out), joined on the data service URI.
Access comes from the service's own access rights in the catalogue; when those are blank (most map services), from Geonorge's kartkatalog record for the service, then from the access rights of the datasets it serves (the strictest wins, and a disagreement between Geonorge and the datasets takes the strictest value). What none of them answer is checked by hand and recorded with the reason in
registry/access-review.yaml. Each entry'ssource.access_fromsays which one decided, and the site shows it. Placeholder links the catalogue holds (example.com) are dropped, and services a manual check found gone are left out throughregistry/retired.yaml.Curate merges hand-written YAML on top: curated fields win, harvested fields fill gaps, matched by catalogue URI or identical endpoint.
Enrich fetches OpenAPI and Swagger documents, including the ones hidden behind a Swagger UI page, and indexes every operation.
Build writes
registry.json(the product),llms.txt, and an SQLite FTS5 index that is rebuilt on demand.Site pre-renders one page per API, provider and category, with sitemap, canonical and Open Graph tags, JSON-LD and a custom 404.
All outbound fetches go through a guard that only talks to public hosts, pins the resolved address, validates every redirect hop and caps body sizes. Design decisions are in docs/adr/.
Entry shape
Field | Meaning |
| stable slug, |
| what the API is; |
| name, orgnr, sector ( |
|
|
|
|
| the links and facts you need to start |
| curated only; |
| curated only; restrictions to respect before relying on the API (terms, undocumented or unstable interface, login for parts of it), shown as «Forbehold» on the site, in |
| OpenAPI summary from |
|
|
| last link-check verdict per URL |
Contribute
There are more open Norwegian APIs than this registry knows about. Three ways in:
No code: suggest an API or report a mistake. Every page on the site has a link that pre-fills the form.
A pull request: add a file in
registry/curated/, runuv run norapi build && uv run norapi check --curated-only && uv run pytest -q, and open the PR. The quality bar is in CONTRIBUTING.md.Register it at data.norge.no if you are the provider; the weekly harvest picks it up from there.
Layout
src/norapi/
harvest/fdk.py Felles datakatalog: search API + SPARQL, raw snapshot to data/raw/
harvest/geonorge.py kartkatalog access fields for services the catalogue leaves blank
curated.py registry/curated/*.yaml and the merge rules
enrich.py OpenAPI resolution and operation extraction
netguard.py public-host guard, address pinning, capped bodies
build.py registry.json, llms.txt, FTS5 index
search.py ranked search with filters, operation search
check.py link checker
samples.py the front page's example calls: run once a week, trimmed for display
site.py static site generator
cli.py / mcp_server.py
registry/ curated YAML, access-review.yaml, retired.yaml, registry.json, llms.txt, specs.json, operations.json, samples.json
site/assets/ design tokens, stylesheet, self-hosted fonts (Geist, OFL), client-side search and console, Open Graph image
tests/ unit tests on recorded fixtures, no network
docs/adr/ decisionsRoadmap
More curated entries, and English summaries for all of them.
Harvest more of Geonorge's kartkatalog (licences, descriptions) for richer geodata entries.
Weekly link checks for every entry, published on the site.
Semantic search next to FTS5 once the registry outgrows keyword matching.
Data sources and terms
Harvested metadata comes from Felles datakatalog, operated by Digitaliseringsdirektoratet, with access constraints for map services from Geonorge's kartkatalog (Kartverket); each entry links back to its catalogue page. The registry holds descriptions and links, not the APIs' data. Every API has its own terms and licence, and the registry only reports what the catalogue or the curator wrote down. Nothing is scraped from agency websites (ADR-0001).
Licence
Code and curated descriptions: MIT. Harvested metadata: see Felles datakatalog.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
This server cannot be deployed
Maintenance
Related MCP Connectors
Turn any task into the right API calls: discover, evaluate, and integrate public APIs.
Norwegian public data for AI agents: companies, tenders, addresses, laws, exchange rates.
Find and call verified APIs from any AI assistant, with free trial calls and previewed writes.
Free, reviewed pilot registry of reusable API knowledge and recipes for AI agents.
Related MCP Servers
- AlicenseBqualityDmaintenanceEnables AI assistants to query Norwegian tax data, submit reports, look up population register information, and validate VAT returns through Skatteetaten APIs using Maskinporten authentication.52MIT
- AlicenseAqualityDmaintenanceEnables AI agents to search and retrieve Norwegian company data from the Brønnøysund Register Centre's open API, including company details, roles, and subunits.5MIT
- AlicenseAqualityDmaintenanceEnables AI agents to access Norwegian public transport and geospatial data, including trip planning, stop departures, address search, elevation, and administrative boundaries, via the Entur and Kartverket APIs.2138 npmMIT
- AlicenseAqualityBmaintenanceThe company registry MCP: company data for AI agents, any country. One MCP server and REST API over national business registries — Norway (Brønnøysundregistrene / Enhetsregisteret, lookup by organisasjonsnummer/orgnr, VAT status) and the United Kingdom (Companies House, lookup by company number, accounts and confirmation-statement deadlines). Five tools: lookup_company, search_company, company_dea560 PyPIMIT