Vehicle Safety Recalls by Make, Model & Year

Check automotive inventory by make, model and year against published NHTSA recall campaigns, with structured records and direct campaign links.

What it returns

One row per distinct campaign/component pair from the official NHTSA vehicle recall endpoint. The default dataset contains these fields:

Field	Type	Meaning
make	string	Make reported by NHTSA
model	string	Model reported by NHTSA
modelYear	integer	Requested and verified model year
manufacturer	string or null	Manufacturer, when provided
campaignNumber	string	NHTSA recall campaign identifier
component	string or null	Affected component, when provided
reportDate	string or null	Report-received date as YYYY-MM-DD; the source's DD/MM/YYYY dates are parsed day first
affectedUnits	integer or null	Campaign affected-unit count, only when the endpoint provides it; never inferred from prose
sourceUrl	string	NHTSA campaign search link

This actor uses automotive NHTSA recall data. FDA drug, device and food recall datasets are outside its scope. It accepts bounded vehicle names and numeric controls; it has no VIN, plate, owner, complaint, document, contact or free-text search inputs. Summaries, consequences, remedies and contact text are discarded from the result projection.

Input

The Console prefill is:

{"make":"Toyota","model":"Camry","modelYear":2020,"maxItems":10}

Input	Required	Limits
make	yes	One NHTSA vehicle make; 1-64 characters
model	yes	One exact NHTSA model name; 1-64 characters
modelYear	yes	Integer from 1949 through the next calendar year
maxItems	no	Integer 1-1000; default 10; cap on returned and charged records

Vehicle names allow letters, digits, spaces and ordinary name punctuation. URLs, line breaks, arbitrary additional keys and invalid numeric values are rejected before the data request. Prefill helps the Console form; API callers must supply the three required vehicle fields.

Real output example

The following record comes from the local Toyota/Camry/2020 run on 2026-10-06 UTC. sample_output.json contains 3 actual records. That run took 2.016 seconds including local process startup; this is an observation, not a performance guarantee.

{
  "make": "TOYOTA",
  "model": "CAMRY",
  "modelYear": 2020,
  "manufacturer": "Toyota Motor Engineering & Manufacturing",
  "campaignNumber": "20V682000",
  "component": "FUEL SYSTEM, GASOLINE:DELIVERY:FUEL PUMP",
  "reportDate": "2020-11-04",
  "affectedUnits": null,
  "sourceUrl": "https://www.nhtsa.gov/recalls?nhtsaId=20V682000"
}

Pricing

Proposed pay-per-event price: $0.0005 per recall record, or $0.50 per 1,000 records in recall events. The event name is recall. A successful dataset write is followed by one recall charge. Duplicates, malformed records, empty results and records beyond maxItems have no recall event. The actor also stops when the configured recall event budget is exhausted.

The price is proposed, not a live Store configuration. Platform usage and platform-managed fees depend on the eventual configuration. Local runs do not make cloud billing transactions. See PRICING.md for the dated Store API evidence and the billing-unit differences between competitors.

Limits and failure handling

Each run checks one make/model/year combination and requests the official endpoint with concurrency limited to one connection. Network errors, HTTP 429 and selected temporary server errors get at most three attempts with 1-second and 2-second backoff. NHTSA also returns HTTP 400 with its explicit successful empty envelope; that exact Count=0, success-message, empty-results shape is treated as no matching records. Other HTTP failures and invalid response envelopes fail the run explicitly; they are not converted into zero recalls.

A malformed individual source record is skipped while valid records continue. Per-item error rows are stored separately in the default key-value store's ERRORS record, preserving the nine-field recall dataset. RUN_SUMMARY records source count, delivered count, item error count, truncation and elapsed time. If all returned records are malformed, the run fails with diagnostics.

The records apply to a make/model/year, not a specific inventory vehicle or its repair status. An empty result can also mean an unlisted or misspelled model. A campaign's unit count, when available, can cover multiple models; it is not your inventory count. The sample endpoint response did not include affected-unit counts, so the sample values are null.

FAQ

Do I need a source API key or proxy? The actor calls the public NHTSA endpoint directly. Its local run used no source API key or proxy.

Does this tell me whether a particular vehicle still needs repair? It lists published campaigns for the requested make/model/year. It does not determine vehicle-specific applicability or completed repairs.

Why is affectedUnits null? The endpoint did not supply a recognized unit-count field. Counts are never calculated from the summary or from the number of recall results.

What if there are more recalls than maxItems? The actor returns the first records in the source's order after validation and duplicate removal. RUN_SUMMARY marks truncation.

What is billed if no records are returned? This actor emits no recall event. Any separately configured platform fees follow Apify's configuration.

Use with AI agents / MCP

After the owner publishes or otherwise makes the actor available, an agent can invoke it through Apify's Actors integration or MCP connection using the four input fields above. The output schema exposes the default dataset items URL. Agents should inspect run status and RUN_SUMMARY before treating an empty dataset as a no-results answer, and use sourceUrl to trace each campaign. This delivery does not implement a separate MCP server.

Local development

The shared project venv already supplies apify, httpx and pytest. Do not install packages into it. From this actor directory, the delivered helpers run the same Python module used by the container:

C:/GritWork/earn/.venv/Scripts/python.exe evidence/run_tests.py unit
C:/GritWork/earn/.venv/Scripts/python.exe evidence/run_tests.py live
C:/GritWork/earn/.venv/Scripts/python.exe evidence/run_local.py
apify validate-schema

The local helper sets APIFY_LOCAL_STORAGE_DIR to this actor's storage directory, writes storage/key_value_stores/default/INPUT.json, and runs python -m src without the Apify CLI. It also sets the SDK 4 storage alias CRAWLEE_STORAGE_DIR to that directory. Unit tests use a saved real NHTSA response; positive/malformed unit-count cases are explicitly synthetic mutations. The marked live test contacts NHTSA. evidence/CONTRACT.md defines the independent process checks.