North Dakota Code MCP server

@pipeworx/north-dakota-code

North Dakota Century Code — state statutes by citation and by official catchline (topic search). Keyless.

Part of Pipeworx — an MCP gateway connecting AI agents to 1764+ live data sources. This is an independent, unofficial integration — not affiliated with, endorsed by, or published by the upstream provider.

Tools

by citation ("12.1-16-01" is murder; "47-16-13.1" is a landlord's duty to maintain a rental unit). Resolves the citation to its one chapter PDF, fetches it live, and extracts the text.

• nd_statute(citation) — full text of a North Dakota Century Code section

own official section and chapter catchlines (not full statutory prose — see below), returning matching citations.

• nd_search(query, limit?) — topic/keyword search over the Legislature's

Auth

Keyless.

The survey trap did not fully apply here

The recorded trap for this pack family was "published only as PDF" — true for the statutory TEXT (ndlegis.gov serves each chapter as a small PDF, cencode/t12-1c16.pdf), but the Legislature also publishes one combined HTML page listing every title, chapter, and section with its official catchline:

https://ndlegis.gov/general-information/north-dakota-century-code/index.html

That page is ~9.5 MB and covers the WHOLE Code (~29,000 sections, 1,740 chapters, 74 titles) in a single document — too large to fetch per call, and there is no lighter per-title listing page (unlike Iowa's) and no search API for the Code (ndlegis.gov's bill-tracking side has a JSON API; the Code side does not, confirmed live). So it is fetched and baked ONCE into src/nd-index-data.ts — citation + official catchline + chapter/title context, NOT statutory text — by scripts/bake-index.mjs, the same "baked table" shape mcps/symmap uses for its herb table. ndsearch reads that baked table entirely in memory; ndstatute uses it only to resolve a citation to its one chapter PDF before fetching and extracting that PDF live, so the statutory text itself is never baked or redistributed.

Refresh path

Re-run the bake script whenever North Dakota recodifies (new session laws land continuously; the index page updates on its own schedule):

node mcps/north-dakota-code/scripts/bake-index.mjs > mcps/north-dakota-code/src/nd-index-data.ts

It fails loudly (not silently) if the page shrinks below ~1 MB or parses to suspiciously few rows — both would mean ndlegis.gov changed its markup rather than that the Code got smaller.

pdf-text.ts needed two real extensions for this state

shared/src/pdf-text.ts (built for Kentucky's per-statute PDFs) didn't handle North Dakota's chapter PDFs out of the box — fixed in a separate commit, with a test:

(/Length 12 0 R) rather than inlining the literal integer. The original detection regex was a backtracking trap for a multi-digit referenced object number (it would silently resolve to the literal integer 1 instead of following the reference) — rewritten as an explicit optional group with nothing to backtrack.

• Every content stream writes /Length as an INDIRECT reference

/Encoding at all — unreadable as WinAnsi bytes. Its own /ToUnicode CMap (keyed by a 1-byte code, unlike Type0's 2-byte CID) is the only place the real character is recorded, so the extractor now checks for that whenever a simple font isn't declared WinAnsi.

• The body text font is a simple (non-Type0) TrueType subset with no

R), which Kentucky's PDFs never used — resolved the same way /Pages already was.

• /Font itself is an indirect reference inside Resources (/Font 18 0

Sectioning without reliable whitespace

Shortened. The full README is on GitHub.