Tn Dmv MCP server

@pipeworx/tn-dmv

Tennessee Department of Safety and Homeland Security: the 186 driver service centers with their per-location service capabilities, plus the state's impaired-driving crash file. Keyless.

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

Why Tennessee gets its own pack

This came out of splitting a single multiplexed us-dmv tool into one pack per state agency. Tennessee's grain is a per-counter capability matrix — REAL ID, CDL skills test, handgun permit, vision test — published on the department's own ArcGIS, which has nothing in common with California's ZIP-level registration snapshot. A union schema across states leaves most arguments ignored on every call, so Tennessee's arguments are Tennessee's.

Tools

tndmvdriverservicecenters filters: city, county, zip (prefix), name, service, limit. tndmvimpaireddrivingcrashes filters: substance, severity, agegroup, gender, agency, groupby, limit.

The distinctive fields

REAL ID, CDL skills test (only 4 of 186 locations do it), handgun permit (47 locations) and vision test are the filters worth knowing about — they are the questions a plain address list cannot answer. office_type separates the 44 full-service centers from 93 self-service kiosks, 46 county-clerk partners and 3 express counters, which matters because a kiosk performs only a handful of the listed services.

Auth

None. tnmap.tn.gov is a public state ArcGIS host.

Gotchas worth knowing

so city is matched after stripping both sides to letters and digits — a caller typing "Oak Ridge" gets the Oak Ridge center rather than a not-found.

• City names run together. USER_City is published as OakRidge, MtJuliet and similar,

round-trip is needed.

• X is longitude and Y is latitude, both as attribute columns, so no geometry

are pre-rendered -separated markup that duplicates columns already returned and roughly triples the payload; with them included the full-layer fetch failed outright rather than merely running slowly. The query names its fields explicitly to leave them behind.

• Two HTML blob columns had to be excluded. USERWebsiteName and USERWebsiteServices

layer 0 holds one row per person involved: ~1,543 person records across ~603 distinct crashes. Both numbers are reported (personrecords, distinctcrashes) because quoting the row count as a crash count would overstate it by 2.5×. It covers injury and fatal crashes; property-damage-only crashes sit outside it. The whole file is calendar year 2022, reported as as_of.

• The crash layer is person-level, and 2022 only. SAFETY/DUICrashesArrestsAgeGender

THP District 3 - Nashville and seven other districts, so agency: "THP" selects all of them; agency: "HIGHWAY PATROL" matches nothing.

• State troopers are not called "Highway Patrol" in the data. They appear as

recorded, and it is the largest bucket (386 of 1,543). It is left in rather than silently dropped.

• "Unknown" is a real age group, the agency's own value for records where age was never

Both are cartographic multi-layer map services (severity-split display layers, basemap boundaries, a weighted-likelihood raster) rather than one queryable table, so they are left out; the DUI layer was the one that returned real, aggregatable records.

• The wider SAFETY/ ArcGIS folder also carries DeerCrashes2019to2023 and I24_CrashHotspots.

Data sources

Shortened. The full README is on GitHub.