Docs

One entry per vulnerability, merged from public sources. Connect an agent over MCP or call the REST API. Both read the same data.

On this page

Connect an agent

The MCP server speaks Streamable HTTP and needs no account, up to the rate limit. Connect your account for the higher one.

Works with most coding agents

npx add-mcp https://everycve.com/mcp --name everycve

claude mcp add --transport http everycve https://everycve.com/mcp

codex mcp add everycve --url https://everycve.com/mcp

Any client that supports Streamable HTTP

https://everycve.com/mcp

Paste this into your coding agent and it does the setup for you.

Setup prompt

Add the Every CVE MCP server to your MCP configuration.

Name: everycve
URL: https://everycve.com/mcp
Transport: Streamable HTTP
Authentication: none needed, under a lower rate limit

If I want the higher limit of an account, use https://everycve.com/mcp/account instead. It uses standard MCP OAuth, so I sign in from my browser when the client asks me to. If this client doesn't support MCP OAuth, keep the first URL and send my Every CVE API key in an "Authorization: Bearer" header.

Use your usual way of adding MCP servers. Then confirm it works by calling get_vulnerabilities with CVE-2014-0160, and tell me if I need to restart anything.

Connect your account

Without an account, MCP allows 120 requests a minute. An account allows 1000, shared by its API key and everything connected to it, so connecting more clients doesn’t add more.

Sign in from your client. Point a client that supports MCP authorization at this URL instead. It’s the same server, but it answers a client with no token with a 401, which is what makes clients open a browser. You sign in with GitHub or email and approve the client. It gets its own short-lived tokens, and never your API key. See and disconnect clients on your account. Replacing your API key disconnects them too.

Account URL

https://everycve.com/mcp/account

/mcp accepts these tokens too, but because it answers anonymous requests, clients have no reason to ask you to sign in there.

Or send your API key. For clients without MCP authorization, keep /mcp and add your API key as a header. X-API-Key also works.

Request header

Authorization: Bearer everycve_…

A key or token that isn’t valid gets a 401 rather than falling back to the anonymous limit. Its WWW-Authenticate header points to the metadata below, so a client can sign in again.

/.well-known/oauth-protected-resource/mcp: Protected resource metadata. /mcp/account has its own at the same path with that suffix.
/.well-known/oauth-authorization-server: Authorization server metadata: the authorization code flow with PKCE S256, for public clients.
Client identity: Client ID Metadata Documents, or Dynamic Client Registration at /oauth/register. Redirects must use HTTPS or a loopback address. Other schemes, such as a custom app scheme, are refused, as the MCP specification requires.
Tokens: Access tokens last an hour and only work at the MCP server, not the REST API. Refresh tokens rotate on each use, and last 30 days unused. Using an old one disconnects the client.

MCP tools

get_vulnerabilities: Look up to 100 entries by any ID. Compact unless detail is full.
check_package: What affects a package, and at a given version. Each match carries evaluated, fixed_in and the sources’ raw claims. truncated means matches were left out.
search_vulnerabilities: Find entries by keyword and filters. A query that is exactly an ID returns every entry that ID names, with the filters and paging applied.
list_changes: What changed since a time.
get_history: How one entry changed, with its EPSS trend. changes_truncated or epss_truncated means older rows were left out.

REST API

Every route is under the base URL below and returns JSON. Requests work without a key, up to the rate limit.

Base URL

https://everycve.com/api/v1

GET`/entries/{id}`

One entry by any ID it's known by: CVE, GHSA, OSV or EUVD. The response includes each source's own values under sources. See the entry object.

Request

curl https://everycve.com/api/v1/entries/CVE-2022-22965

Response, shortened

{
  "id": "CVE-2022-22965",
  "aliases": ["GHSA-36p3-wjmg-h94x"],
  "status": "published",
  "title": "Spring Framework JDK 9+ Remote Code Execution Vulnerability",
  "severity": {
    "version": "3.1",
    "score": 9.8,
    "vector": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
    "level": "critical",
    "assessor": "primary",
    "from": "nvd"
  },
  "epss": {
    "day": "2026-11-01",
    "score": 0.99638,
    "percentile": 0.9995,
    "model": "v2026.06.15"
  },
  "facts": {
    "in_kev": true,
    "public_exploit": true,
    "fix_available": true
  },
  "cwes": ["CWE-94"],
  "published": "2022-04-01T22:17:30Z",
  "exploits": [
    {
      "source": "metasploit",
      "id": "exploit/multi/http/spring_framework_rce_spring4shell",
      "type": "exploit",
      "date": "2022-03-31",
      "url": "https://www.rapid7.com/db/modules/exploit/multi/http/spring_framework_rce_spring4shell"
    }
  ],
  "affected": [
    {
      "source": "ghsa",
      "ecosystem": "Maven",
      "package": "org.springframework:spring-beans",
      "ranges": [
        {
          "type": "ECOSYSTEM",
          "events": [
            { "introduced": "5.3.0" },
            { "fixed": "5.3.18" }
          ]
        }
      ]
    }
  ]
}

POST`/entries/batch`

Up to 100 entries in one request. An ID with no entry is left out of the response.

Request

curl https://everycve.com/api/v1/entries/batch \
  -d '{"ids": ["CVE-2023-4863", "GHSA-j7hp-h8jx-5ppr"]}'

Response, shortened

{
  "entries": [{ "id": "CVE-2023-4863", "status": "published" }]
}

GET`/entries`

Search by keyword and filter. Results leave out each entry's heaviest fields. Pass next as offset to continue.

Search parameters
Parameter	Meaning
`q`	Keywords, or an exact ID. An exact ID returns the entry it names, and every other entry its advisory covers when it names several CVEs. The other parameters filter and page those entries like any search
`kev`	In CISA KEV: `true` or `false`
`epss_gte`	EPSS score at least, 0 to 1
`cvss_gte`	CVSS score at least, 0 to 10
`cwe`	A weakness, such as `CWE-79`
`published_after`	A date, as `2026-01-31`
`published_before`	A date, as `2026-01-31`
`vendor`	Vendor name, best effort
`product`	Product name, best effort
`cpe`	A CPE 2.3 string
`status`	`published`, `disputed`, `withdrawn` or `rejected`
`limit`	Up to 100, default 25
`offset`	Up to 5000

Request

curl 'https://everycve.com/api/v1/entries?q=openssl&kev=true&epss_gte=0.5'

Response, shortened

{
  "entries": [
    {
      "id": "CVE-2014-0160",
      "status": "published",
      "severity": {
        "version": "3.1",
        "score": 7.5,
        "level": "high",
        "from": "nvd"
      },
      "facts": {
        "in_kev": true,
        "public_exploit": true,
        "fix_available": true
      }
    }
  ],
  "next": 1
}

GET`/packages`

What affects a package. Pass ecosystem and name, and optionally version to keep only what affects that version. Withdrawn and rejected entries are left out.

With a version, a match is left out only when every source's claim rules that version out. If any claim says it is affected, affected is true. Otherwise a claim that couldn't be compared, such as git commits or a range the source wrote in a form we can't read, leaves evaluated false and affected null, which means the version may or may not be affected. It never means safe.

fixed_in: Versions the sources report as fixing it. A fix beside an unreadable range is the source's own claim.
claims: Each source's raw affected record for the package: its ranges and any explicit versions. A range of type UNPARSED keeps the source's own text in raw, with at most a fixed event. Read these by hand when evaluated is false. With a version, claims that ruled it out are left out.
truncated: true when the result hit its limits and some matches were left out.

Request

curl 'https://everycve.com/api/v1/packages?ecosystem=npm&name=lodash&version=4.17.20'

Response, shortened

{
  "ecosystem": "npm",
  "name": "lodash",
  "version": "4.17.20",
  "matches": [
    {
      "entry": { "id": "CVE-2021-23337", "status": "published" },
      "affected": true,
      "evaluated": true,
      "fixed_in": ["4.17.21"],
      "claims": [
        {
          "source": "ghsa",
          "ecosystem": "npm",
          "package": "lodash",
          "ranges": [
            {
              "type": "ECOSYSTEM",
              "events": [{ "introduced": "0" }, { "fixed": "4.17.21" }]
            }
          ]
        }
      ]
    }
  ]
}

GET`/entries/{id}/history`

How one entry changed since Every CVE first saw it, plus its daily EPSS scores, newest first. Each list is capped at its newest rows: 500 changes and 3000 days of scores.

changes_truncated: true when older changes exist beyond those listed. Absent otherwise.
model: The EPSS model that made each score. Scores jump when the model changes, so compare scores only within one model.
epss_truncated: true when older scores exist beyond those listed. Absent otherwise.

Request

curl https://everycve.com/api/v1/entries/CVE-2014-0160/history

Response, shortened

{
  "id": "CVE-2014-0160",
  "changes": [
    {
      "entry": "CVE-2014-0160",
      "seq": 36,
      "at": "2026-11-01T18:40:13Z",
      "sources": ["pocgithub"],
      "fields": ["exploits"]
    }
  ],
  "epss": [
    { "day": "2026-11-01", "score": 0.99999, "percentile": 0.99996, "model": "v2026.06.15" }
  ]
}

GET`/entries/{id}/exploits`

Every exploit link for one entry. An entry keeps at most 25 links from each source, and exploits_more on the entry counts the rest. This lists them all, in the entry's order: links to public code first, then the oldest.

Request

curl https://everycve.com/api/v1/entries/CVE-2021-44228/exploits

Response, shortened

{
  "exploits": [
    {
      "source": "pocgithub",
      "id": "tangxiaofeng7/CVE-2021-44228-Apache-Log4j-Rce",
      "title": "tangxiaofeng7/CVE-2021-44228-Apache-Log4j-Rce",
      "date": "2021-12-09",
      "url": "https://github.com/tangxiaofeng7/CVE-2021-44228-Apache-Log4j-Rce"
    }
  ]
}

GET`/changes`

Everything that changed since a time. Pass since and optionally limit, then pass next as after to continue.

Request

curl 'https://everycve.com/api/v1/changes?since=2026-11-01T00:00:00Z'

Response, shortened

{
  "changes": [
    {
      "entry": "CVE-2014-0160",
      "seq": 1,
      "at": "2026-11-01T18:40:13Z",
      "sources": ["cve"],
      "fields": ["created"]
    }
  ],
  "next": 1
}

GET`/sources`

Each source's sync state and license. The sources page shows the same data.

status: How the last attempt went: ok, error or never.
overdue: true when the source has gone too long without a successful sync, or never had one. status only describes the last attempt, so a source can be ok and overdue. Treat a source as healthy only when status is ok and overdue is false.

Request

curl https://everycve.com/api/v1/sources

Response, shortened

{
  "sources": [
    {
      "name": "cve",
      "title": "CVE List",
      "kind": "records",
      "about": "The official CVE records, and the starting point for most entries.",
      "homepage": "https://www.cve.org",
      "license": "CVE Terms of Use",
      "interval": "30m",
      "last_success": "2026-11-01T18:40:13Z",
      "status": "ok",
      "records": 312408,
      "overdue": false
    }
  ]
}

GET`/stats`

The headline counts. They’re saved after each ingest that changes an entry, and one row a day is kept, so this reads them rather than counting anything.

known_cves: Entries with a CVE ID, leaving out rejected and withdrawn ones.
exploited: Entries in CISA KEV.
recent: Entries published in the last recent_days days.
recent_trend: up or down against the span before it, or empty when level.
updated_at: When the counts were taken, or null before the first count.

Request

curl https://everycve.com/api/v1/stats

Response

{
  "known_cves": 312408,
  "exploited": 1484,
  "recent": 4120,
  "recent_days": 30,
  "recent_trend": "up",
  "updated_at": "2026-11-01T18:00:00Z"
}

GET`/feed.xml`

The 100 newest items for a feed reader: entries as they’re first seen, and entries as they’re added to CISA KEV. It needs no key.

The feed sits at the site’s root, not under the API’s base URL.

Format: Atom 1.0, served as application/atom+xml. This is the only feed format for now. For JSON, use recent changes.
Item IDs: The entry’s URL, ending in #new or #kev, so one vulnerability can appear once for each.
Refresh: Cached for up to 5 minutes.

Feed URL

https://everycve.com/feed.xml

Response, shortened

<feed xmlns="http://www.w3.org/2005/Atom">
  <title>Every CVE: new vulnerabilities and KEV additions</title>
  <id>https://everycve.com/feed.xml</id>
  <updated>2026-11-02T03:46:46Z</updated>
  <entry>
    <title>CVE-2023-4863 added to CISA KEV: Google Chromium WebP Heap-Based Buffer Overflow Vulnerability</title>
    <id>https://everycve.com/CVE-2023-4863#kev</id>
    <updated>2023-09-13T00:00:00Z</updated>
    <link href="https://everycve.com/CVE-2023-4863"/>
  </entry>
</feed>

GET`/legal`

The privacy policy and terms of service, the same text the website shows.

updated: The day the document last changed.
markdown: The document’s body as Markdown, without its title.

Request

curl https://everycve.com/api/v1/legal

Response, shortened

{
  "privacy": {
    "updated": "2026-11-02",
    "markdown": "This policy explains what \"Every CVE\" (the \"Service\") collects…"
  },
  "terms": {
    "updated": "2026-11-01",
    "markdown": "These terms cover \"Every CVE\" (the \"Service\")…"
  }
}

The entry object

An entry is one vulnerability, merged from every source that mentions it. Top-level fields hold the value Every CVE chose, and sources keeps what each source said. An advisory covering several CVEs belongs to each CVE's entry and is listed in also.

Fields of an entry
Field	Type	Meaning
`id`	string	The primary ID, a CVE where one exists
`aliases`	string[]	Other IDs for the same vulnerability
`status`	string	`published`, `disputed`, `withdrawn` or `rejected`
`severity`	object	The lead CVSS score. The CNA wins, then NVD, then others. `from` names the winner
`epss`	object	Latest chance of exploitation in the next 30 days. `model` names the EPSS model that made the score, and scores from different models aren't comparable. The first model is `v1`
`facts`	object	`in_kev`, `public_exploit` and `fix_available`. `public_exploit` is true only when a source links to code that exploits the vulnerability. A KEV listing alone doesn't set it, nor does a detection template such as Nuclei's, nor a Metasploit module that is not an exploit, such as an auxiliary scanner
`kev`	object	The CISA KEV listing, when there is one
`exploits`	object[]	Links to public exploit code, and to detection templates and scanners that only check for the flaw. Every CVE stores no exploit code. At most 25 per source, links to public code and the oldest first
`exploits_more`	number	Exploit links left out of `exploits`. The exploits route lists them all. Absent when none are
`affected`	object[]	Packages, version ranges and explicit versions, by source. A range of type `UNPARSED` keeps the source's own text in `raw` and can only mean unknown
`products`	object[]	Vendor and product pairs
`references`	object[]	Links about the vulnerability
`sources`	object	Each source's own records, on single-entry reads
`also`	string[]	Other entries an ID resolved to

The public GET /api/v1/config endpoint reports the current rate limits and request caps. These pages use the same values.

Rate limits and keys

120requests a minute with no key

1000requests a minute with an API key

Send a key as a bearer token. The limit belongs to the account, so replacing a key or connecting another client doesn’t start a new count. Every response reports your remaining requests in X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset.

Request

curl -H "Authorization: Bearer everycve_…" \
  https://everycve.com/api/v1/entries/CVE-2024-3094

Errors

Errors share one shape. fields appears when specific parameters failed.

404 response

{
  "error": { "code": "not_found", "message": "no entry for CVE-2099-1" }
}

Connect an agent

Connect your account

MCP tools

REST API

GET/entries/{id}

POST/entries/batch

GET/entries

GET/packages

GET/entries/{id}/history

GET/entries/{id}/exploits

GET/changes

GET/sources

GET/stats

GET/feed.xml

GET/legal

The entry object

Rate limits and keys

Errors

GET`/entries/{id}`

POST`/entries/batch`

GET`/entries`

GET`/packages`

GET`/entries/{id}/history`

GET`/entries/{id}/exploits`

GET`/changes`

GET`/sources`

GET`/stats`

GET`/feed.xml`

GET`/legal`