Fda 510k Medical Devices Recalls Enforcement Scraper

Find medical device applicants, regulatory contacts, and classified product recalls directly from official FDA records.

Build MedTech prospect lists from 510(k) decisions or monitor enforcement recalls across medical devices, drugs, and food. Receive flat, exportable JSON with company names, published contacts, product descriptions, classifications, dates, addresses, and source links. Requests use the official openFDA API with no browser, proxy, or API key configuration.

Best for tracking 510(k) decisions and classified enforcement recalls. Choose one of the four supported datasets and a recent date range. Warning letters are outside this Actor's coverage, and source updates can lag the underlying event.

Ready-to-run example

The Input form includes public example values. A fresh apify run uses the input schema defaults. To explicitly select the small sample, including when input was saved by an earlier local run, use:

apify run --input-file .actor/input.example.json

You can also paste .actor/input.example.json into the JSON input editor in Apify Console. Replace the example sources with your own targets and increase the limits when you are ready.

How to use

1. Choose a dataset: device 510(k) clearances, device recalls, drug recalls, or food recalls.
2. Select an inclusive date range. Optionally add a company/product phrase or company state code.
3. Set the maximum number of saved records and run the Actor. The prefilled example saves 3 records.
4. Download JSON, CSV, or Excel from the dataset, or connect its API to your workflow.

Results are requested newest first, 100 records per API page. Pagination is sequential to respect the public API's rate limits and avoid fetching unnecessary pages. A run can save up to 5,000 records on a paid Apify plan, subject to the event budget and matching data.

What data can you extract?

Mode	Official source	Included data
DEVICE_510K	device/510k.json	K number, applicant, published regulatory contact, device, specialty, FDA decision, decision date, address
DEVICE_RECALLS	device/enforcement.json	Recall number, recalling firm, device description, hazard class, status, recall reason, report date, address
DRUG_RECALLS	drug/enforcement.json	Drug recall number, recalling firm, description, hazard class, status, reason, report date, address
FOOD_RECALLS	food/enforcement.json	Food recall number, recalling firm, description, hazard class, status, reason, report date, address

Each record contains exactly 18 flat fields. status describes extraction (SUCCESS or ERROR); decisionOrRecallStatus contains the FDA decision or recall status. Missing source values are empty strings. contactPersonName comes only from a published contact field, typically in 510(k) records. Enforcement records generally do not contain a contact person.

actionDate is the decision date for 510(k) records and the enforcement report date for recalls, normalized to YYYY-MM-DD. A recall's report date may differ from its initiation date. reasonForRecallOrSummary contains the recall reason, or the FDA's summary/statement availability indicator for 510(k) records; it does not download a summary PDF. medicalSpecialtyOrClassification contains the device specialty or the recall hazard classification, such as Class I or Class II.

This Actor queries the four endpoints above. Warning letters, inspection findings, PMA approvals, and summary PDFs are outside these datasets. The 510(k) dataset can include different decision codes; inspect decisionOrRecallStatus when selecting prospects. The historical local folder name does not change the available modes.

Input settings

Setting	Default	Details
datasetMode	DEVICE_510K	One of the four modes above
searchKeyword	Empty	Literal phrase, up to 200 characters, searched across applicant/product/contact for 510(k), or firm/product/reason for recalls
startDate	2026-01-01	Inclusive decision/report date; real calendar date required
endDate	2026-10-15	Inclusive end date, on or after start date
stateFilter	Empty	Exact FDA company address state/province code, normalized to uppercase; does not filter recall distribution
maxItems	100	Integer 1–5,000; prefill 3; includes successful and error rows
timeoutSecs	15	Integer 3–60; timeout per HTTP attempt

{
  "datasetMode": "DEVICE_510K",
  "startDate": "2026-01-01",
  "endDate": "2026-10-15",
  "maxItems": 3,
  "timeoutSecs": 15
}

Keyword and state values are escaped as literal Lucene phrases. Wildcards and advanced query expressions are not executed. Empty filters are omitted. Keep the keyword empty for the broadest sample; there is no guaranteed match for an arbitrary phrase, date range, or state.

Output example

The following illustrative record shows the output shape; it is not a claim about a real FDA clearance.

{
  "recordType": "DEVICE_510K",
  "status": "SUCCESS",
  "referenceNumber": "K260001",
  "companyName": "Example Medical Devices Inc.",
  "contactPersonName": "Jordan Smith",
  "productOrDeviceName": "Example Monitoring System",
  "medicalSpecialtyOrClassification": "Cardiovascular",
  "decisionOrRecallStatus": "Substantially Equivalent",
  "reasonForRecallOrSummary": "Summary",
  "actionDate": "2026-02-05",
  "streetAddress": "100 Example Avenue, Suite 200",
  "city": "Boston",
  "state": "MA",
  "zipCode": "02110",
  "country": "US",
  "fdaSourceUrl": "https://api.fda.gov/device/510k.json?search=k_number.exact%3A%22K260001%22&limit=1",
  "scrapedAt": "2026-10-20T12:00:00.000Z",
  "errorMessage": ""
}

All strings are sanitized for spreadsheet export, including removal of null bytes and protection against formula injection. Numeric values and valid phone-like strings are preserved. Consumers displaying these fields in a webpage must render them as text or apply their normal context-specific output encoding.

Use the results in Clay, n8n, Make, or your API

Use Apify's integration or HTTP modules in Clay, n8n, and Make to run the Actor with the JSON input above. Retrieve the default dataset after completion and filter on status = SUCCESS. Use referenceNumber together with recordType as the key for upserts across scheduled runs. Recalls can change status over time, so update existing rows rather than assuming each reference is immutable.

To read a completed dataset, request GET https://api.apify.com/v2/datasets/{datasetId}/items?format=json&clean=true using the dataset ID from the run. Use your integration's secure credential configuration when authentication is required. A run-completion webhook can trigger enrichment, CRM import, Slack alerts, or a compliance review workflow. This Actor itself does not send messages or perform contact enrichment.

Pricing and limits

The custom pay-per-event name is fda-record-scraped. Only successfully saved SUCCESS rows receive that custom event. ERROR rows cost $0 in custom event fees, and a valid query with no matches saves no rows and charges no custom events. Apify infrastructure/storage or other platform fees may apply separately.

For a concrete example, at a configured event price of $0.001 per successful record, 1,000 successful records cost $1.00 in event fees. A run saving 900 successful records and 100 error records costs $0.90 in event fees at that rate. These examples illustrate the calculation; the active Store pricing determines the actual rate. Before publishing, configure the event and price in Apify Console. Repository metadata does not activate monetization.

Hosted runs for non-paying Apify users have a hard cap of 25 saved rows, counting both success and error rows. Local runs do not have this Free plan cap. Both the input limit and the remaining event budget are enforced before saving. When the event budget is exhausted, extraction stops cleanly. LIMITED_PERMISSIONS is supported; the Actor never queries /v2/users/me.

No proxies are used. The openFDA access limits document 240 requests/minute and 1,000 requests/day per IP without a key; limits and shared datacenter IP quotas can affect availability. Requests are spaced at least 300 ms apart. Transient 429/5xx and network failures receive at most two retries with bounded backoff. A 403 is reported safely without attempting proxy rotation. JSON responses are capped at 2 MiB after decompression. Network errors become uncharged error rows instead of failing the run; only invalid user input exits with code 1.

FAQ and troubleshooting

Why are there no results? Check the dates, selected mode, literal keyword, and company state code. An openFDA NOT_FOUND response means the query has no matches and is handled as a successful empty run.

Why is the contact name empty? The FDA may not publish a contact. The Actor does not infer names or enrich them from third-party sources.

Does this include every recall announcement? It uses classified FDA enforcement reports. Dataset publication timing and coverage follow the official openFDA enforcement documentation. It does not monitor press releases or replace the underlying FDA record.

Can I search a company's full name? Yes. Enter the phrase as plain text. Keywords search multiple relevant fields, and punctuation is escaped. Narrow phrases may return fewer matches than a single distinctive word.

What do error rows mean? errorMessage contains a safe failure category such as HTTP_429, ETIMEDOUT, INVALID_JSON, or RESPONSE_TOO_LARGE. A page failure stops further pagination while preserving already saved records. No raw response, internal path, stack trace, or credential is written to an error row.

What happens on cancellation? Aborting, SIGTERM, and SIGINT cancel in-flight requests and flush buffered rows. The writer serializes dataset writes and enforces the remaining billing budget during shutdown as well.

Can I request more than 5,000 rows? Split work across narrower date windows and deduplicate by mode and reference number. This keeps each run within its configured result and event budget. See the official query parameters and 510(k) field reference for source semantics.

Local development and deployment

Use Node.js 22 or later. Run npm ci, npm run build, and npm test. Run npm run smoke for live, read-only checks of all four FDA endpoints with the default date window. The smoke checks do not push datasets or bill events. Run npm run smoke:entrypoint to verify the actual SDK entrypoint in isolated temporary local storage, including all four modes, a no-match query, input rejection, exact row counts, and process memory. These local runs do not bill platform events. Tests use node:test and injected transports for deterministic security, streaming, retry, pagination, and billing coverage. Child runners set UV_THREADPOOL_SIZE=64 and NODE_ENV=production.

For a local Actor run, save the example JSON to storage/key_value_stores/default/INPUT.json, then run npm start. Run from a directory that does not contain environment files. Production deployment uses the included multi-stage node:22-alpine Dockerfile, a non-root user, 256 MiB default memory, and a 3,600-second run timeout. The seven input settings are intentionally limited to the FDA use case; page fetching is sequential and has no browser or proxy configuration.