norway-archives-mcp

An MCP server for Norwegian family history: the National Archives' church books, censuses and emigrant lists and their scans, the National Library's books and newspapers, and today's official place names.

• Digitalarkivet, run by Nasjonalarkivet (the National Archives of Norway), holds the parish registers (ministerialbøker and klokkerbøker), the censuses of 1801–1920, emigrant protocols, probate records and much more. Some of it is transcribed and searchable; far more is scanned and can only be read page by page. The server does both: it searches the transcriptions, maps a scanned book's lists (baptisms, confirmations, marriages, burials, in- and out-migration) to image numbers, and downloads a page with its permanent id.
• Nasjonalbiblioteket (the National Library) has digitised nearly every Norwegian book and newspaper, including the bygdebøker, the farm-by-farm local histories genealogists live in. Many open only from Norwegian addresses, but a search still tells you which page a name is on.
• Kartverket's place-name register gives today's spelling, type and coordinates of a farm or hamlet.

It works the way a careful genealogist does. The scanned page is the evidence. A transcription is a lead (most church-book transcriptions were keyed by contractors for Ancestry, MyHeritage and FamilySearch), and a search that finds nothing has only searched what was transcribed. Every page comes back with its source, archive reference, image number and permanent link, so the citation points at the archive, not at this server.

Nothing here writes anywhere but your own disk, and nothing here keeps a family tree.

Tools

The server publishes thirteen tools. All but the two downloads are read-only; those create one new file each and never overwrite one.

A typical path through an untranscribed list (confirmations, say):

1. da_search_sources(municipality="Røros", list_type="kf") finds the church books that hold confirmations, and says the confirmations are scanned but not searchable.
2. da_get_source(source_id=6540) gives the image the confirmations start on.
3. da_list_pages(source_id=6540, section="kf") lists those pages with their permanent ids.
4. da_download_page(page_id="kb2007…", destination=…) saves each page to read, with a citation such as Church book from Røros parish 1860-1878 (1640P), AV/SAT-A-1456/681/L0932, image 171 of 464 (kb20070419660172). Nasjonalarkivet (The National Archives of Norway), Digitalarkivet. https://goto.digitalarkivet.no/kb20070419660172.

Setup

You need Python 3.11 or later and uv. There is no key to request.

Without cloning. uvx fetches it from PyPI and runs it in one step:

uvx norway-archives-mcp

From a clone, which is what you want if you will change it:

git clone https://github.com/ianderso/norway-archives-mcp
cd norway-archives-mcp
uv sync
uv run norway-archives-mcp   # stdio server, usually launched by the client

Either way the server speaks MCP over stdio, so you will normally let an MCP client start it rather than run it by hand.

Claude Desktop

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

A desktop app does not always inherit your shell's PATH. If the server fails to start because uvx cannot be found, give the full path that which uvx prints as the command.

Claude Code

claude mcp add norway-archives -- uvx norway-archives-mcp

Configuration

Nothing is required. A .env file in the directory the server starts in supplies anything the environment does not; only that directory is read.

An unusable value is reported on the first tool call as a not_configured result naming the variable.

Being a good guest, and what the terms say

Digitalarkivet publishes no API and no terms-of-use page; its footer links only privacy and accessibility statements. Its reuse page says scans without restrictions "kan disse fritt brukes videre både i trykksaker og på internett": they may be freely reused in print and online, with the holding archive credited (digitalarkivet.no/content/1570). Its robots.txt asks every crawler for five seconds between requests and disallows a list of named bots, AI agents among them. This server is not a crawler: it fetches the pages a researcher asks for, one at a time, at least five seconds apart (configurable upward only), within a daily budget, with a User-Agent that names it. Search, record and source pages are cached as parsed results for 1, 30 and 30 days; manifests for 30; permalink lookups for a year.

The National Library's API is documented and keyless (api.nb.no). Access to each item follows the library's own rules: material marked for Norway opens only from Norwegian addresses (foreign researchers can apply for access to the Bokhylla collection), and this server reports that rather than working around it. One request a second.

Kartverket's place-name API needs no login or authorisation, in its own words "ikke nødvendig med innlogging/autorisasjon" (api.kartverket.no/stedsnavn). Two requests a second at most.

A 429, a 5xx or a dropped connection gets one retry, honouring Retry-After. A bot check is reported as blocked and left alone.

How to read what comes back

• A transcription is not the record. Read the scanned page (da_download_page) before citing a name, a date or a relationship.
• A zero is narrow. da_search_persons searches transcribed lists only. da_search_sources shows which lists are scanned but not searchable; confirmations and migration lists often are. Page through them.
• Municipalities moved. The da_ tools filter by the historical municipalities Digitalarkivet uses (see municipalities); place_names reports today's, after the mergers of 2020 and earlier.
• Section boundaries are approximate. da_get_source gives the image a list starts on; it usually runs until the next list starts, but books were not always kept in order.
• Norway-only books still answer. nb_search_inside gives the printed page number of a hit even where the page cannot be shown here.
• Census pages (ids beginning ft) come from the older viewer at full resolution only.

Deliberately not here

• Writing anything, or logging in. No tool sends a correction, a report or a message to any service.
• The Norwegian Historical Population Register (HBR), whose search is a form and whose API is for partners.
• Norway Heritage, whose terms forbid duplication or redistribution without permission.
• Working around access rules: Norway-only material, restricted recent records, or bot checks.

Security

Tool arguments are written by a model, and the model reads text this server does not control: record transcriptions, catalogue entries, OCR. The server assumes that text can steer the model, and limits what a steered model can make it do.

• Which hosts. Requests go only to the three services' fixed hosts over https. Any other host is refused, including addresses taken from a response. Redirects are followed only within one service.
• How much. A page or JSON answer over 8 MB, or an image over 60 MB, is refused as it streams in.
• Which files. The downloads create one new file and never overwrite one. The bytes must be a JPEG, judged by their first bytes, and the file must end in .jpg or .jpeg. Never a hidden file or folder, never under ~/Library, and with NORWAY_ARCHIVES_DOWNLOAD_DIR set, never outside it, all judged after links are resolved. A refused download leaves nothing on disk.
• Arguments are validated (page ids, item ids, record addresses) before they reach a request.
• Site text is untrusted. The server's instructions tell the model to treat record and catalogue text as material to weigh, never as instructions.

To report a vulnerability, see SECURITY.md.

Development

uv sync --extra dev
uv run pytest                      # mocked with respx; never touches a site
uv run ruff check .
uv run ruff format --check .
uv run python -m tests.live_check  # paced calls to the live services

See CONTRIBUTING.md for how the suite is organised, docs/API-NOTES.md for what was observed of each service and when, and docs/DESIGN.md for why the server is shaped this way.

Credits

The records, scans and catalogue belong to Nasjonalarkivet, Nasjonalbiblioteket and Kartverket. Most church-book transcriptions were made by AMF (Ancestry, MyHeritage and FamilySearch) under agreement with the National Archives.

License

MIT.