lovspor

Ask an AI assistant about Norwegian law and you get a confident answer — statute name,
section number, sometimes a quote. Nothing in that loop checks whether the section
actually exists. Lovspor does.

Lovspor keeps an open, daily-updated copy of Norwegian law — close to 6,000 acts and
central regulations from Lovdata's public data, with full
version history — and gives any AI assistant (Claude Desktop, Claude Code, Cursor, any
MCP client) tools to search it, quote it, and verify citations against the real text.

What you can ask

Once connected, your assistant answers from the live corpus instead of stale training data:

• "What changed in skatteloven this year?"
• "What did husleieloven § 9-6 say in May 2026?" — full version history since April 2026, diffable between any two tracked dates (fetch the corpus with --full-history, see below)
• "Does this paragraph exist?" — validate_citation answers instead of guessing
• "Is this quote verbatim?" — verify_quote checks it against the actual text

Quickstart

Requires uv. No clone, no account, no API key.

# 1. Fetch the legal corpus to ~/.cache/lovverk (re-run any time to update)
uvx lovspor fetch-corpus

# 2. Connect your AI client — Claude Code:
claude mcp add lovverk -- uvx lovspor mcp

Other MCP clients (Claude Desktop's claude_desktop_config.json, etc.):

{
  "mcpServers": {
    "lovverk": {
      "command": "uvx",
      "args": ["lovspor", "mcp"]
    }
  }
}

Restart the client and lovverk appears in its MCP list. Seventeen of the eighteen tools
work immediately — no key, and no network access beyond your local corpus clone.

The point-in-time tools need the full history. fetch-corpus clones shallow
(--depth 1) by default, which keeps the download small and is enough for every
current-law tool. On a shallow clone, get_law_at and diff_law_versions reach only as
far back as the clone does. Fetch it whole instead:

uvx lovspor fetch-corpus --full-history

The same flag deepens an existing shallow clone in place (git fetch --unshallow), so this
is not a decision you have to get right the first time.

Full setup guide, all eighteen tools with examples, troubleshooting and limitations:
docs/mcp.md.

What's inside — and what's not

Inside: all current Norwegian acts (lover) and central regulations (sentrale
forskrifter) from Lovdata's public-data API, re-synced daily at 04:00 UTC, each with a
structured per-act change history. Live count: the corpus_status tool.

Not inside: court decisions, preparatory works (forarbeider), agency circulars
(rundskriv), and municipal regulations beyond a small, opt-in set of local regulations
observed on the authorities' own websites (ADR-0016; served only by qualified address or
dataset="lokale-forskrifter", labelled asserted: false — see docs/mcp.md). A rule can
be binding and absent here — an empty result is not evidence that no such rule exists.

Optional: search by meaning

semantic_search is the one tool that needs an embedding key (OpenAI; the corpus
vectors ship pre-computed — only your query is embedded). Bring your own key via the
server's env:

{
  "mcpServers": {
    "lovverk": {
      "command": "uvx",
      "args": ["lovspor", "mcp"],
      "env": { "OPENAI_API_KEY": "sk-...your-own-key..." }
    }
  }
}

It's your key in your own local config file — keep that file private and never commit
it. Without a key, semantic_search is simply disabled; the other seventeen tools are
unaffected. Details: docs/embeddings.md.

Optional: hosted endpoint

Don't want to self-host? A hosted MCP endpoint runs at
https://lovspor.no/mcp. Chat-app connectors (Claude.ai, ChatGPT) log in
through OAuth; developer clients (Claude Code, Cursor) use a bearer token —
ask for one. See docs/mcp.md for both,
or to run the same thing yourself.

How it works

A scheduled workflow pulls Lovdata's public-data tarballs daily, classifies each
document as new / updated / renamed / removed, renders deterministic Markdown, and
pushes the diff to lovverk — the
public corpus repo — as conventional-commit history. lovspor (this repo, on
PyPI) is the engine and MCP server; legal text
never lives here.

Architecture and design rationale: docs/decisions.md.
Release process: docs/releasing.md.
Offline evals for the MCP surface: evals/ (repo-only tooling).

Install from source

git clone https://github.com/bartoszkobylinski/lovspor
cd lovspor
./scripts/bootstrap.sh     # uv sync + pre-commit hooks
uv run lovspor --help

Sources

• https://api.lovdata.no/v1/publicData/get/gjeldende-lover.tar.bz2 — current Norwegian laws
• https://api.lovdata.no/v1/publicData/get/gjeldende-sentrale-forskrifter.tar.bz2 — current central regulations

Data is licensed under Norsk lisens for offentlige data (NLOD) 2.0.

License

The engine code in this repository is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). See LICENSE. Versions up to and including commit 632dae8 were published under MIT; everything after is AGPL-3.0. The corpus data in lovverk remains NLOD 2.0 and is unaffected.

The legal text produced by this engine is published in the
lovverk repository under NLOD 2.0,
with attribution to Lovdata.

Related work

• cloveras/lovdata2 — JSON tooling and MCP
  server for the same Lovdata public data. lovspor is complementary, focused on
  Markdown rendering, Git-based change tracking, and an MCP server scoped to the
  lovverk corpus shape.