NVD CVE Search with Source Evidence

Search the official NVD vulnerability API by keyword, CVE ID, product CPE, CVSS v3 severity, and publication date. Export one normalized row per CVE plus a coverage report and source evidence for each delivered page.

This product uses data from the NVD API but is not endorsed or certified by the NVD.

The NVD API itself is free. This Actor packages pagination, date windows, normalized exports, source provenance, and explicit delivery limits for scheduled vulnerability research and data pipelines.

Start a search

Omitted input uses keywordSearch: "log4j", pubStartDate: "2026-01-01", keywordExactMatch: false, and maxResults: 100. The end date is the current UTC date. The default is a fixed publication-date query, not a moving update monitor.

For historical Log4Shell records:

{"keywordSearch":"log4j","pubStartDate":"2021-12-01","pubEndDate":"2021-12-31","maxResults":100}

For a direct lookup:

{"cveId":"CVE-2021-44228"}

A CVE ID takes precedence over keyword, date, CPE, severity, and exact-match controls. Ignored controls are listed in the report.

For a product and severity query, clear the keyword explicitly:

{"keywordSearch":"","cpeName":"cpe:2.3:a:apache:log4j","cvssV3Severity":"CRITICAL","pubStartDate":"2021-12-01","pubEndDate":"2021-12-31","maxResults":100}

Inputs

Input	Meaning
keywordSearch	NVD English-description keyword query. Empty string clears the default.
cveId	One CVE-YYYY-NNNN... identifier. Uses direct ID lookup.
cpeName	CPE 2.3 product prefix or exact version. Product prefixes use the official virtualMatchString filter; specified versions use cpeName, padded with wildcard attributes when abbreviated.
cvssV3Severity	LOW, MEDIUM, HIGH, or CRITICAL. This is the NVD v3 filter; output scoring follows the selection policy below.
pubStartDate	Inclusive YYYY-MM-DD; default 2026-01-01. Empty string disables the date filter.
pubEndDate	Inclusive YYYY-MM-DD; default current UTC date when a start date exists. For an end-only range starting at 1970-01-01, explicitly set pubStartDate: ""; omitting it retains the default start date.
keywordExactMatch	Exact phrase flag; requires a nonempty keyword. Default false.
maxResults	Maximum distinct delivered CVEs, 1–10,000. Default 100.
batchSize	Records requested per source page, 1–2,000. Default 100.
maxRequests	Maximum attempted source requests. Default 1,000; maximum 10,000.
maxRunSeconds	Stop admitting new requests or data writes after this time budget. Default 3,600; maximum 7,200. Platform run timeouts also apply.

Dates use full calendar-day bounds. Longer ranges are divided into contiguous windows of at most 120 calendar days. Pages advance by source offset. Requests wait at least 6.5 seconds after the previous response, without a key or proxy. There are no automatic source retries.

Dataset and evidence

Every row includes these 18 fields:

Fields	Meaning
cveId, description, status	CVE identity, English description, and NVD analysis status.
published, lastModified	Source calendar dates as YYYY-MM-DD. Full timestamps remain in page evidence.
cvssScore, severity, cvssVector	Score, severity, and vector from the same selected assessment.
attackVector, attackComplexity, privilegesRequired, userInteraction	CVSS v3.1 dimensions; null for a selected v2 assessment.
exploitabilityScore, impactScore	Original selected subscores, without clamping to a different CVSS version's range.
weaknesses	Stable, deduplicated CWE codes, excluding NVD placeholder codes.
references	First ten source reference URLs, preserving order and duplicate positions. No linked content is fetched.
sourceIdentifier, extractedAt	Source organization and collection time in UTC.

Select Primary CVSS v3.1, otherwise the first v3.1 assessment; then fall back to Primary v2 or first v2. CVSS v3.0/v4 assessments do not replace this policy. Their presence is disclosed in provenance. Missing optional values are null; explicitly empty collections remain empty. V2 access/authentication values are kept in provenance and are not relabeled as v3 dimensions. A missing score does not mean a vulnerability is harmless.

OUTPUT and RUN_REPORT contain coverage status, stop reason, acknowledged delivery count, source totals and drift, severity/publication-year summaries, billing bookkeeping, and page evidence keys. source-page-NNNNN artifacts contain observed rows, raw-body and canonical-record SHA-256 hashes, scoring version/source, and full source timestamps. An evidence artifact records observations and intended delivery; it is not proof that the Dataset write completed.

Export Dataset rows through Apify's JSON, CSV, or Excel downloads and API. Platform schedules, webhooks, and integrations can repeat the same input. This Actor has no persistent update watermark or asset-exposure assessment.

Limits, failures, and fees

$0.001 per delivered CVE ($1 per 1,000), with zero startup, empty-result, duplicate, or separate Dataset event fee. Use Apify's maximum total charge setting to cap event fees; source requests are bounded separately. Platform run usage is included in this tariff. Downloading, exporting, retaining, or accessing data outside the run may have separate platform charges under your Apify plan.

complete: true means all requested date windows and pages were exhausted. A result, request, fee, or time limit that leaves requested coverage unfinished produces complete: false with a reason. If the last page exhausts the last requested window exactly at the result cap, complete: true is still correct. A valid empty NVD response is distinct from 403, 429, timeout, invalid JSON, or changed source structure. Source failures fail the run and retain earlier acknowledged rows. If a repeated CVE changes during a scan, collection stops and reports the changed ID instead of silently discarding its update.

If a storage or charge acknowledgment is unknown, the Actor stops without replaying that mutation. Partial writes and SDK bookkeeping must be reconciled against the Dataset and platform run charges. Existing run reports prevent automatic resumption within the same run. A fresh run is a separate scan and may incur new charges.

Collection is not an atomic database snapshot. NVD counts and records can change during pagination. CPE matches do not establish that a particular asset/version is affected; reference links are not verified patch or exploit evidence. Responses must be JSON with identity encoding and at most 16 MiB. Redirects are rejected. No API key, login, arbitrary source URL, or advisory downloads are accepted.

Cloud runs require a finite platform timeout. Source requests and pacing stop admitting work at that deadline, including during any platform shutdown window; earlier Dataset rows remain saved. The runtime rejects disabled or enlarged Python integer-parsing limits and disables redirects for SDK writes, so a redirected write is never automatically replayed.

Official references: NVD vulnerability API, NVD developer guidance, NIST reuse notices.