Nj Mvc MCP server

@pipeworx/nj-mvc

New Jersey DMV data: Motor Vehicle Commission (MVC) inspection and emission-repair facilities, and the record-level vehicle-inspection results file. Keyless.

Part of Pipeworx — an MCP gateway connecting AI agents to 1476+ live data sources.

What New Jersey calls its DMV

New Jersey's agency is the Motor Vehicle Commission, universally the MVC. Residents say "MVC agency" where other states say "DMV office", so the tool descriptions name both.

Tools

Auth

None. data.nj.gov is a public Socrata portal; the public rate limit handles these queries without an app token.

The distinctive dataset

gvur-kt7q is 2,314,032 per-test records, each carrying make, model, model year, fuel, full OBD-II readiness monitors, and separate emissions and safety verdicts. No other state publishes inspection outcomes at that grain, which is what makes "which cars fail New Jersey emissions inspection most" answerable at all.

Live-verified statewide numbers: 2,080,550 passes and 233,477 failures — about a 10% overall fail rate. Split by test type, initial tests fail 8.7% of the time and re-tests 26.7%.

mintests (default 100) exists because a fail-rate ranking without it is meaningless — a model with three tests and two failures would top every table. sort="failrate" ranks by failure share; the default ranks by volume.

Gotchas worth knowing

These are live-verified behaviours, not guesses:

WATERFORD WORK address — Waterford Works is in Camden County — is labelled MORRIS. The county filter is kept because agents ask for counties, but prefer city when the answer has to be right, and the response says so in its note.

• The facility file's county column contradicts the address on some rows. A

renew a New Jersey licence needs an MVC agency, which this file does not contain. The tool description is explicit so a router does not send that question here.

• These are inspection and repair shops, not licensing agencies. Someone asking where to

partial city string matches more reliably than a full one.

• Town names are abbreviated and upper case (WATERFORD WORK, not "Waterford Works"), so a

null, so latitude/longitude come back null for those.

• Coordinates are missing on some facility rows — geocoded_column is absent rather than

test_type="R" isolates the re-tests.

• Inspection rows are tests, not vehicles. A car that fails and returns is counted twice;

which is the overall verdict column.

• emissionsfailures and safetyfailures can overlap and do not have to sum to failed,

them out of any ranking by construction.

• Makes and models are upper case and abbreviated (TOYOTA, CAMRY, F150).
• Model years include junk — a handful of rows carry 2033 and 2041. min_tests filters

offered as a groupby. njabody_style is the working equivalent.

• njavhcltype is empty on every row despite being documented as P/T/H, so it is not

a Socrata throttle, and the metadata helper does not retry, so a sequential call intermittently returned a null refresh date.

• asof is fetched concurrently with the data**, not after it. The 2.3M-row aggregate earns

reported in as_of rather than assumed current.

• The inspection file was last refreshed 2026-03-12; the facility file 2026-07-20. Both are

Data sources

Shortened. The full README is on GitHub.