CVE Vulnerability Lookup — NVD Search & AI Triage

Turn any product, library or keyword into a prioritised vulnerability report. Give this Actor a name like log4j, openssl or a specific id like CVE-2021-44228, and it searches the official NVD (National Vulnerability Database) for every known CVE — then hands back each one with its CVSS score and severity, affected products, CWE weaknesses and references, plus an AI triage that tells you which vulnerabilities actually matter.

You bring the product name. It hands back a ranked risk brief.

---

The flow

1. Give a product / library / CVE     2. Search the official NVD API      3. Get a triaged risk brief
   (e.g. "log4j", "openssl",     →        (CVSS, CWE, affected,       →      CVE rows + AI triage:
    "CVE-2021-44228")                       references)                       top priority · themes · risk

1. Add your queries — product names, library names, keywords, or exact CVE ids. One per line.
2. The Actor searches NVD 2.0 (keywordSearch, or cveId when the query is a CVE id), pulling CVSS v3.1/v3.0/v2 scores, CWE weaknesses, affected CPEs and references.
3. The AI layer triages each query's CVE set — what to fix first, common weakness patterns, an overall risk level, and concrete recommended actions.

---

What you get per query

For every query, the Actor:

1. Searches the NVD database — the official U.S. government source of record for CVEs. CVE ids are detected automatically (/^CVE-\d{4}-\d+$/) and looked up by id; everything else is a keyword search.
2. Returns one record per CVE — cveId, published, cvssScore, cvssSeverity, cvssVersion (prefers 3.1, falls back to 3.0 then 2.0), English description, cwe (weakness ids), affectedProducts (best-effort from CPE configurations) and the top references.
3. Filters by severity — set minCvssScore to drop everything below your floor (e.g. only 7.0+ high/critical).
4. Scopes by date — set publishedAfter to only see recent CVEs.

Then the AI layer (OpenAI gpt-4o-mini, a fast and low-cost model) adds one triage record per query:

• overallRisk — low / medium / high / critical, reflecting the most severe exploitable issues present.
• topPriority — the CVEs that matter most, each with whyItMatters.
• themes — common weakness patterns across the set (e.g. deserialization, path traversal).
• summary — a two-sentence read on the risk.
• recommendedActions — concrete next steps. Everything is grounded strictly in the CVE data — no invented CVE ids or scores.

---

Three ways people use it

• Dependency & supply-chain security audits — point a CVE search at every library in your stack (log4j, openssl, lodash, spring) and get a triaged list of known vulnerabilities with CVSS scores — a lightweight vulnerability scanner for your dependency tree and software bill of materials (SBOM), straight off the NVD API.
• Vendor / vulnerability due diligence — before adopting a tool or acquiring a company, search its products for a CVSS-ranked history of CVEs and let the AI triage flag the supply chain security red flags that matter.
• Threat monitoring — track a technology by keyword with publishedAfter to surface newly published CVEs, with an AI read on severity and recommended actions.

Keywords this Actor serves: CVE search, NVD API, vulnerability scanner, CVSS, supply chain security, CWE, security due diligence, dependency audit.

Rate limits: The free NVD API allows roughly 5 requests per 30 seconds without a key, so the Actor paces its requests and keeps concurrency low. Need more throughput? Set an optional NVD_API_KEY environment variable (a free NVD key) and the Actor sends it as the apiKey header. It works fine without a key.

---

Input

Field	Type	Notes
queries	array	Required. Products, libraries, keywords, or CVE ids (e.g. log4j, openssl, CVE-2021-44228). One per line.
maxCvesPerQuery	integer	Max CVEs to fetch from NVD per query. Default 20 (max 2000).
minCvssScore	number	Only keep CVEs with a CVSS base score ≥ this floor (0–10). Default 0.
publishedAfter	string	Optional ISO date (e.g. 2023-01-01) — only CVEs published on/after it.
maxResults	integer	Global cap on total CVE records across all queries. Default 100.
maxConcurrency	integer	Queries run in parallel. Default 2 — NVD rate-limits hard.
model	string	Override the OpenAI model. Default gpt-4o-mini.

Minimal input example

{
  "queries": ["log4j", "openssl"],
  "maxCvesPerQuery": 20,
  "minCvssScore": 7.0
}

---

Output example

A CVE row:

{
  "recordType": "cve",
  "query": "log4j",
  "cveId": "CVE-2021-44228",
  "published": "2021-12-10T10:15:09.143",
  "cvssScore": 10.0,
  "cvssSeverity": "CRITICAL",
  "cvssVersion": "3.1",
  "description": "Apache Log4j2 2.0-beta9 through 2.15.0 ... JNDI features used in configuration, log messages, and parameters do not protect against attacker controlled LDAP and other JNDI related endpoints...",
  "cwe": ["CWE-502", "CWE-917"],
  "affectedProducts": ["apache log4j", "oracle retail xstore point of service"],
  "references": [
    "https://logging.apache.org/log4j/2.x/security.html",
    "https://www.cisa.gov/uscert/ncas/current-activity/2021/12/10/apache-log4j-vulnerability-guidance"
  ]
}

A triage row (one per query):

{
  "recordType": "triage",
  "query": "log4j",
  "cveCount": 20,
  "overallRisk": "critical",
  "topPriority": [
    { "cveId": "CVE-2021-44228", "whyItMatters": "Unauthenticated remote code execution (Log4Shell), CVSS 10.0, trivially exploitable and widely weaponised." }
  ],
  "themes": ["Unsafe deserialization / JNDI lookups", "Denial of service via recursive lookups"],
  "summary": "The Log4j CVE set is dominated by the critical Log4Shell RCE and its follow-on patches. Any service embedding a pre-2.17 Log4j 2.x build is at severe, actively-exploited risk.",
  "recommendedActions": [
    "Upgrade Log4j to 2.17.1 or later across all services.",
    "Scan deployed artifacts and containers for vulnerable log4j-core jars."
  ]
}

---

Pay-per-event pricing

This Actor uses pay-per-event billing with one event:

Event	Price	Charged when
query-analyzed	$0.15	Once per product/keyword searched against the NVD database and triaged with AI.

The Actor logs its OpenAI token usage and the estimated AI cost per query at the end of every run, so the event price always stays safely above the model cost.

---

Data source

All vulnerability data comes from the official NVD — National Vulnerability Database (NIST), via the public NVD API 2.0. NVD data is free and public; this Actor adds rate-limit-safe fetching, a clean flat schema and an AI triage layer on top.

---

Setup

The AI triage calls the OpenAI API. Set the environment variable / secret OPENAI_API_KEY on the Actor.

Optionally set NVD_API_KEY (a free key from NVD) to raise the rate limit — not required.

---

Use it from an AI agent (Apify MCP server)

This Actor is available through the Apify MCP server, so AI agents (Claude, ChatGPT, or anything that speaks MCP) can call it as a tool — hand it a list of products or CVE ids, and read back the triaged vulnerabilities — with no custom integration code.

---

Run it locally

npm install
# put your input in storage/key_value_stores/default/INPUT.json
export OPENAI_API_KEY=sk-...   # required for the AI triage
# export NVD_API_KEY=...       # optional, higher rate limit
apify run -p

---

Related actors

• npm Package Intelligence — Analyze any npm package — health, maintenance, dependencies and risk signals.
• PyPI Package Intelligence — The same for Python packages on PyPI — metadata, releases and risk.
• GitHub Repo Analyzer — Profile any GitHub repository — activity, maintainers and project health.
• B2B Tech Stack Detector — Detect the technologies a company's website runs on, for sales and research.