Skip to main content
Menu

Agents

Everything on cve.now is free to read by machine: no key, no account. An optional cve.now API key gives your own rate limit instead of the shared one.

MCP

Add the MCP server

Streamable HTTP, no auth. Two tools: search finds an operation and execute runs it. The old tools (get_vulnerability, list_feed and the rest) still answer; they are deprecated and leave these docs on 2027-01-10.

Set up the cve.now MCP server for me. cve.now is a free, public CVE feed: no login, no API key.

- Server name: cve-now
- Transport: Streamable HTTP
- URL: https://mcp.cve.now/mcp

Add it the way you normally add MCP servers (your CLI command or your MCP config file), then confirm it connects and lists its two tools, search and execute. To finish, use them to run the cve.get operation on one recent CVE and summarize it for me.

Paste this into any coding agent. It adds the server with its own MCP setup.

entities.search
Search cve.now's whole catalog of vulnerabilities and early warnings
entities.get
One entity by id, any type
entities.list
One page of one entity type
feed.read
The items a /feed subscriber gets, as JSON Feed, newest first
rankings.list
Every ranking the site has, with its id, title, the entity type it ranks, its periods
rankings.get
One ranking's rows, ranked
feedback.send
Send a private note to the cve.now team
stack.check
Check exact package versions against cve.now's synced vulnerability catalog
cve.list
The newest CVEs and advisories, narrowed by priority slice, vendor, technology, ecosystem or fix status
cve.get
One CVE by id
advisory.list
The newest advisories
advisory.get
One advisory by id
early-warning.list
Early warnings
early-warning.get
One early warning by id: an SZB- id (a breach shares its SZB- id; use breach.get for one)
breach.list
Breaches and ransomware attacks, newest first
breach.get
One breach by id: an SZB- id
vendor.list
The vendors cve.now follows, with their CVE counts
vendor.get
One vendor by id: the vendor's slug
package.list
The packages that have advisories
package.get
One package by id: <ecosystem>/<name>, such as npm/lodash or npm/@babel/core
tech.list
The technologies cve.now follows
tech.get
One tech by id: the technology's slug
leaderboard.get
Get the cve.now Index

REST

REST API

Every MCP operation has a REST twin that takes the same inputs and returns the same JSON.

OpenAPI
Get vulnerability by CVE, GHSA or OSV id
GET /api/v1/entries/<id>
Read vulnerability feed
GET /api/v1/entries?name=<feed> (or tech, ecosystem, vendor, fix)
Check dependencies
POST /api/v1/check
Send private feedback to the cve.now team
POST /api/v1/feedback
Get worst-CVE leaderboard (the cve.now Index)
GET /api/v1/leaderboard
Search vulnerabilities
GET /api/v1/search
Get a package's vulnerability history
GET /api/v1/packages/<ecosystem>/<name>
Every entry, newest first
GET /api/v1/entries
Upload a lockfile for stack.check or a stack feed
POST /api/v1/stacks
Sync and coverage status
GET /api/v1/status
Public usage counts
GET /api/v1/stats
curl -L https://cve.now/api/v1/entries/CVE-2021-44228

Feeds

Feeds

RSS by default, or Atom and JSON Feed. Pick the types (CVEs, early warnings, breaches), then narrow by filter (critical, kev), technology or package ecosystem, or build a feed of only what affects your lockfile.

https://cve.now/feed?filter=critical&format=json
https://cve.now/feed?tech=nextjs,react
https://cve.now/feed?type=cve,early-warning&filter=critical

Polling

Polling guide

If you re-read cve.now on a schedule, ask only for what changed. On the first run page with since set to a time up to 7 days back; every N minutes after, send the next_since of the last answer and the ETag of the last response as If-None-Match. A 304 has no body. Add ids to follow a watchlist: only the entries on it that changed come back.

https://cve.now/api/v1/entries?since=2026-11-08T00:00:00Z&limit=200
https://cve.now/api/v1/entries?ids=CVE-2024-3094,CVE-2023-4863&since=2026-11-08T00:00:00Z

Limits

Limits

Without a key, a list or search reaches the newest 400 results (two pages of 200). Deeper access will require a free API key, which is coming soon. A limited request answers 429 with Retry-After.

Depth
The newest 400 results of a list or search
Delta sync
since reaches back 7 days with no depth limit; ids takes 25 per request
Search
30 a minute per client
Single entry
120 a minute per client
MCP tool calls
100 in 10 minutes per client

Docs

Docs for agents

Entry, day and browse pages also have a Markdown copy: add .md to the URL, or send Accept: text/markdown.

llms.txt · llms-full.txt · openapi.json · Entry schema · agent.json · mcp.json · Agent skill

Something wrong or missing? Run the feedback.send operation, POST /api/v1/feedback, or the form on the About page.

Changelog

Latest changes

Dates are UTC. The full changelog is also JSON and an RSS feed.

Full changelog

AddedAPI · MCP · Web · CLI

Malicious packages are advisories, and check fails on them

OSV's MAL- reports and GitHub's malware advisories are now advisories on cve.now. /api/v1/check, the MCP check_stack tool and cve.now check report an installed malicious release as a critical finding with malicious: true, and the CLI exits non-zero on it.

  • /api/v1/advisories?malicious=true and /feed?type=advisory&malicious=true list them. A malicious advisory's page is noindex and is not in the sitemaps.
  • debug@4.4.2 and chalk@5.6.1 are findings; chalk@5.6.0 is not.
  • CLI 0.4.0 carries the malware handling; the GitHub Action defaults to it.
ChangedMCP

The MCP server returns an Mcp-Session-Id

initialize now answers with an Mcp-Session-Id header, a signed token that holds only the client's name and version. Sending it back changes nothing, and no client needs to.

ChangedWeb

Sign in in a new tab with visible progress

Continue with Sazabi opens sign-in in a new tab while the original page shows progress, offers Cancel and continues automatically after sign-in. If the browser blocks the new tab, sign-in continues in the current tab.

FixedWeb

Sign in with a Sazabi account that has no display name

Signing in with a verified Sazabi account that has no display name now completes. Your verified email is used as the display name until Sazabi supplies one.

ChangedAPI · MCP · Web

Search reads a misspelled word as a vendor, technology or package

GET /api/v1/search, entities.search and /search now read each word of 5 or more characters against vendor, technology and package names, so githb returns GitHub and axiom suggests axios. The page says when it also searched a corrected word.

  • resolved.words lists { word, outcome, source, as } for each such word: exact, corrected (also searched as the name in as), nearby (a real word beside a name) or none. A quoted word is never corrected.
  • items puts an exact or corrected name first, then entries that name the word in a title or entity field, then nearby names, then the other matches.
  • Vendor, technology and package rows in items carry match as { name, kind } and entries, the advisories naming the entity (0 when none).
  • suggestions carries reason: spelling is the query with a misspelled word written as the name it is near, and now comes with results; broader is still sent only when nothing matched.
  • note is sent only when items is empty, and says what was searched and what was not. type:package axios no longer carries it beside the package row.
  • /search shows "Also showing results for" a corrected word, with a link that searches only the typed word, and "Did you mean" for a nearby one. The search box suggests the same names as you type, so githb suggests GitHub.
  • Nothing is renamed or removed; total and entries keep their meaning.