Wrench.Pro

/data

Open vehicle data

Everything on this site is built from public records. This section is where you can see the records themselves: what we hold, where each row came from, how far back it goes, when it was last refreshed, and what we do not have. Download it, check our work, or build something of your own.

Pages under /data show numbers and sources only. What the numbers might mean lives in Research, and every chart there links back here.

Datasets

Known Issues clusters

published

NHTSA complaints for each year, make and model, grouped by failure theme. Every quoted complaint is checked byte-for-byte against the federal record. The only dataset here that does not exist anywhere else.

rows35,307 clusters
vehicles5,822 year/make/model · 56 makes
years1990–2027
sourceNHTSA ODI complaints
data through2026-10-12
licenseCC BY 4.0 (derived) · NHTSA records public domain
bulk.csv26.5 MBbulk.json36.6 MBREADME.txtbrowse by make →
  • NHTSA complaints (summary)published

    Complaint counts per year/make/model with date span, crash and fire flags, and the top five component categories.

    7,924 rows · 7,924 vehicles · 1990–2027 · through 2026-09-20 · csv · json

  • NHTSA recallspublished

    Recall campaigns matched to year/make/model: campaign number, component, summary, consequence, remedy.

    17,944 rows · 4,486 vehicles · 1990–2027 · through 2026-09-20 · csv · json

  • NHTSA investigationspublished

    Defect investigations (PE, EA, RQ, DP) matched to the vehicles they name, with status and any resulting recall.

    4,216 rows · 1,998 vehicles · 1990–2027 · through 2026-09-20 · csv · json

  • FARS occupant deathspublished

    Occupant deaths in fatal crashes per year/make/model from NHTSA's Fatality Analysis Reporting System, with match provenance.

    8,940 rows · 8,940 vehicles · 1990–2025 · through 2026-10-11 · csv · json

  • Maintenance intervals

    Manufacturer service intervals in our normalized fields. Schedules live as hand-maintained modules with editorial notes; the interval-only export needs a JSON form first.

    on site · export planned

Use it from your AI

The same data is live behind an MCP server. Add one URL to Claude, ChatGPT, Gemini, VS Code or any MCP client and your assistant can look up recalls, Known Issues and repair-cost estimates for any US vehicle since 1990. No download, no key, same license.

https://mcp.wrench.pro/mcp
setup for each client →plain JSON API example →

Coverage by model yearfull coverage page →

What we have for each model year, as a share of the vehicles in our catalog. Known Issues clusters are only attempted where a vehicle has at least 30 complaints on file; the last column is clustered vehicles as a share of those. Gaps are gaps. The 1990s will fill in as the backfill runs. Same numbers as our internal ops dashboard, as of 2026-10-12.

yearvehiclesschedulerecalls shownKI eligibleKI clusteredKI %
2027353100%1111100%
2026366100%1563131100%
2025385100%229105105100%
2024389100%244151151100%
2023390100%263153153100%
2022388100%259161161100%
2021379100%262151151100%
2020390100%266156156100%
2019377100%279169169100%
2018379100%278163163100%
2017376100%258184184100%
2016362100%250176176100%
2015356100%270183183100%
2014349100%248180180100%
2013335100%237176176100%
2012330100%235176176100%
2011317100%229179179100%
2010316100%239166166100%
2009327100%5170170100%
2008329100%819519499%
2007315100%8198198100%
2006302100%5185185100%
2005294100%3185185100%
2004297100%9177177100%
2003284100%10178178100%
2002285100%13180180100%
2001274100%15175175100%
2000261100%18170170100%
1999250100%15161161100%
1998236100%13152152100%
1997248100%19147147100%
1996255100%16137137100%
1995270100%17145145100%
1994260100%18132132100%
1993265100%16130130100%
1992262100%20123123100%
1991268100%23117117100%
1990257100%22105105100%

How to read this section

  • Counts are counts. There is no denominator for how many of each vehicle are on the road, so nothing here is a rate, a rank or a grade.
  • Every dataset page states its source, its schema, and its known gaps. See methodology and coverage.
  • Make and model pages under each dataset (for example /data/recalls/toyota) hold the rows as plain tables.
  • Derived data is CC BY 4.0. Link back to the page you used.
  • Prefer a live lookup over a download? The MCP server and the JSON API serve the same rows, refreshed on the same schedule.

Cite this

Derived data is CC BY 4.0. Link back to the page you used. Text and BibTeX below.

Wrench.Pro (2026). Open vehicle data [dataset]. https://www.wrench.pro/data
@misc{wrenchpro_data_2026,
  author = {Wrench.Pro},
  title = {Wrench.Pro open vehicle data},
  year = {2026},
  howpublished = {\url{https://www.wrench.pro/data}},
  note = {Derived from NHTSA public records. CC BY 4.0.}
}

hub · known-issues · methodology · license · research (interpretation) · connect (MCP)