NVD API with KEV and EPSS: how to triage a CVE
Use the NVD API with CISA KEV and EPSS to triage a CVE: CVSS score, exploited status and exploit odds, plus NVD API key rate limits.
Luisa · · 7 min read
On this page
The NVD API gives you a CVSS score, and a CVSS score tells you how bad a vulnerability could be. It does not tell you whether anyone is exploiting it. To triage a CVE from code, you want the NVD record plus two more signals: whether the flaw is on the CISA Known Exploited Vulnerabilities list, and what its EPSS score says about the chance of exploitation. Those signals come from three public sources with three different APIs, which is why many teams end up sorting by severity alone.
This guide runs all three on Log4Shell (CVE-2021-44228), shows the answers, and covers the NVD API key and its rate limits. By the end you can write your own triage function, or hand the job to your agent with the prompt below. looot serves the three sources through one key and one request shape.
What does each source answer?
Each source answers a different question, which is why they work better together than any one does alone. NVD describes the flaw and how severe it could be. KEV records that exploitation has actually been seen. EPSS forecasts how likely exploitation is in the next 30 days.
| Source | Question | Log4Shell, 2026-11-16 | looot endpoint |
|---|---|---|---|
| NVD CVE API | What is it, and how severe? | CVSS 3.1 base score 10.0 | `nvd-cve-get` |
| CISA KEV catalog | Is it known to be exploited? | Listed, added 2021-12-10 | `cisa-kev-catalog` |
| FIRST EPSS API | How likely is exploitation? | 0.99999, percentile 1.0, score date 2026-11-15 | `first-epss-score` |
CISA publishes KEV as one file in JSON and CSV. Version 2026.11.13, released on 2026-11-13, holds 1,781 entries.
How do you get a CVE record from the NVD API?
Send a GET request to the CVE API 2.0 with `cveId`, and you get back the description, CVSS scores, weaknesses and affected configurations. The record also carries the KEV fields when the CVE is on that list, so this one call often answers two of the three questions. Without a key, the request needs nothing else:
curl -s "https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2021-44228"Through looot, the same lookup is a run of `nvd-cve-get`:
curl -s -X POST "https://api.looot.ai/v1/runs?wait=20" \
-H "Authorization: Bearer $LOOOT_TOKEN" \
-H "content-type: application/json" \
-d '{"endpointId":"nvd-cve-get",
"input":{"cve_id":"CVE-2021-44228"},
"idempotencyKey":"triage-CVE-2021-44228-nvd"}'Here are the fields from that answer that matter for triage, trimmed from the full record:
{"vulnerabilities":[{"cve":{"id":"CVE-2021-44228",
"cisaExploitAdd":"2021-12-10",
"metrics":{"cvssMetricV31":[{"cvssData":{"baseScore":10.0}}]}}}]}When `cisaExploitAdd` is missing, the CVE is not on the KEV list. A CVE the NVD does not have comes back as a miss, which you should record and not retry. A miss is free.
Do you need an NVD API key?
You need one only when you send more than a few requests. The NVD's Start Here page sets the public rate limit at 5 requests in a rolling 30 second window, and 50 requests in the same window with an API key. Even with a key, the NVD recommends that your script sleep for six seconds between requests.
| Without a key | With a key | |
|---|---|---|
| Requests per rolling 30 seconds | 5 | 50 |
| Where the key goes | none | request header `apiKey` |
| Pause the NVD recommends | six seconds between requests | six seconds between requests |
To get a key, fill in the form on the Request an API Key page with a valid email address. The NVD emails you a link, and you must activate the key within seven days or request a new one. A key belongs to one requestor, and activating a new key turns off the old one. The 2.0 APIs take the key in the request header:
curl -s -H "apiKey: $NVD_API_KEY" \
"https://services.nvd.nist.gov/rest/json/cves/2.0?cveId=CVE-2021-44228"To keep a full local copy, page through the CVE API with a key and that pause, then refresh no more than once every two hours with a last-modified window, as the Start Here page describes.
How do you check the exploit probability?
Run `first-epss-score` with the same id, or call `https://api.first.org/data/v1/epss?cve=CVE-2021-44228` directly. EPSS answers with the probability, from 0 to 1, that the CVE is exploited in the wild in the next 30 days, along with its percentile among all scored CVEs and the date of the score.
{"data":[{"cve":"CVE-2021-44228","epss":"0.999990000",
"percentile":"1.000000000","date":"2026-11-15"}]}If you have a whole list of CVEs, `first-epss-scores` takes up to 50 in one call, and `first-epss-top` returns the highest scored. When your starting point is a package and not a CVE, `osv-package-advisories` returns its open-source advisories.
How would you triage a CVE with these signals?
Start with KEV, because exploitation that has already been seen outweighs any forecast, so those flaws get fixed first. High EPSS scores come next, since those are likely to be exploited soon, while a high CVSS score with a low EPSS score can usually be scheduled. This order is our suggestion and not a standard, so set thresholds that fit your own systems.
Log4Shell scores at the top of all three, which makes it an easy demo and a poor test. The approach earns its keep on your own backlog, where the signals disagree.
To run it on one CVE, paste this into your agent:
Prompt for your agent
Triage one CVE
The nvd-cve-get catalog page lists its inputs and current price, and the FIRST EPSS provider page lists the three EPSS endpoints. For jobs that have several providers, read how data provider fallback works, and for paid sources see what your agent's data calls cost.
What it cannot do
This does not scan your systems, so you bring the CVE ids or package names yourself. It does not merge the three answers into one record either: your code makes the calls and combines them. The record comes straight from NVD, so if NVD is down the call fails and you try again later.
looot sends NVD calls without an NVD key, so they share the public limit. For a bulk pull of thousands of records, request your own key and call the NVD directly with the pause it recommends. Watching for new CVEs on a timer needs scheduled runs, not available yet; until then, call `nvd-cve-search` from your own scheduler. If you want either of these, ask your agent to send the request to looot as feedback.
This product uses data from the NVD API but is not endorsed or certified by the NVD. See EPSS at https://www.first.org/epss.
Questions
Keep reading
Data provider fallback: automatic fallback for data APIs
How data provider fallback works when a data API returns nothing or fails: the order, the cost cap, what stops a route and who pays for a miss.
The looot team · · 6 min read
AI agent cost: model tokens plus data calls
AI agent cost has two meters, model tokens and data calls. How to estimate each, why the data side is easier to cap, and how to read per-call prices first.
The looot team · · 7 min read
Idempotency keys for paid APIs: never pay twice
How an idempotency key stops a retry from charging you twice on a paid API, how to build one, and how to cap the cost of a call when a retry goes wrong.
Luisa · · 8 min read
Try it with the agent you already use
Start your workspace, top up, and paste one prompt into your agent.