SazabiNot published by vendorVerified 2026-10-28

free CVE API and MCP server for Claude & AI agents

One merged entry per vulnerability from eight public sources, as a JSON API, RSS feeds and a keyless MCP server with a lockfile checker. Free, no login.

Compare Docs

Verified factsChecked 2026-10-28

Price	Free, with no login and no API key. The FAQ says Sazabi already merges this data for its own product, so publishing it costs little.³ [Verified 2026-10-28]
API key / auth	None. The REST endpoints and the MCP server at https://mcp.cve.now/mcp are unauthenticated; our keyless calls on 2026-10-28 returned HTTP 200.² [Verified 2026-10-28]
Rate limit	No numeric limit is published, and our test responses carried no rate-limit headers.⁴ [Verified 2026-10-28]
License / terms	Model-written summaries are CC BY 4.0; upstream data keeps its own license and attribution (NVD data, not endorsed by NVD). No separate terms of service page is published.⁵ [Verified 2026-10-28]
Version status	No beta or GA label. The MCP server reports version 0.1.0, and the site says it launched on Oct 22, 2026.¹ [Verified 2026-10-28]
Sources and freshness	CVE List, GitHub advisories, OSV and KEV are checked every few minutes; NVD and EUVD every ten minutes; EPSS changes a few times a day. /api/v1/status showed all eight sources OK on the check date.³ [Verified 2026-10-28]
Coverage	397,140 vulnerabilities tracked on 2026-10-28 (/api/stats). Package matching covers npm, PyPI, Go, Maven, crates.io, Packagist, RubyGems, Hex and NuGet lockfiles.⁶ [Verified 2026-10-28]
Endpoints	GET /api/v1/entries/<id> (follow redirects), /api/v1/entries (cursor paging, filter=critical,kev), /api/v1/status, /feed with filter, tech and ecosystem, /stack?p=… and POST /api/stack for lockfile upload.² [Verified 2026-10-28]

FeaturesChecked 2026-10-28

YesNoN/ANot published

Capability	Value	Note
StatusNo beta or GA label and no SLA. Launched Oct 22, 2026; the MCP server reports version 0.1.0.	Not published¹ [Not published by vendor: checked 2026-10-28 on https://cve.now/]	No beta or GA label and no SLA. Launched Oct 22, 2026; the MCP server reports version 0.1.0.
API key requiredNo key, no login. REST, feeds and MCP all answered HTTP 200 without auth on 2026-10-28.	No² [Verified 2026-10-28 by test request to https://cve.now/llms.txt]	No key, no login. REST, feeds and MCP all answered HTTP 200 without auth on 2026-10-28.
Free tierFree; no paid tier. The FAQ says it is free with no login or key; Sazabi merges the data for its own product.	Yes³ [Verified 2026-10-28 on https://cve.now/#faq]	Free; no paid tier. The FAQ says it is free with no login or key; Sazabi merges the data for its own product.
Rate limitNo numeric limit is published, and our test responses carried no rate-limit headers.	Not published⁴ [Not published by vendor: checked 2026-10-28 on https://github.com/sazabi/cve.now]	No numeric limit is published, and our test responses carried no rate-limit headers.
Official MCP serverHosted, mcp.cve.now/mcp. Streamable HTTP, no auth; tools get_vulnerability, list_feed and check_stack (tools/list run 2026-10-28).	Yes⁴ [Verified 2026-10-28 on https://github.com/sazabi/cve.now]	Hosted, mcp.cve.now/mcp. Streamable HTTP, no auth; tools get_vulnerability, list_feed and check_stack (tools/list run 2026-10-28).
Official SDKNo SDK is published; the docs describe plain HTTPS endpoints, feeds and MCP.	Not published⁴ [Not published by vendor: checked 2026-10-28 on https://github.com/sazabi/cve.now]	No SDK is published; the docs describe plain HTTPS endpoints, feeds and MCP.
Data license / termsUpstream data keeps its own license and attribution; NVD data, not endorsed by NVD. No terms page.	Summaries CC BY 4.0; sources keep theirs⁴ [Verified 2026-10-28 on https://github.com/sazabi/cve.now]	Upstream data keeps its own license and attribution; NVD data, not endorsed by NVD. No terms page.
Commercial use allowedNo terms of service or commercial-use rule is published; CC BY 4.0 on summaries allows commercial reuse with credit.	Not published⁴ [Not published by vendor: checked 2026-10-28 on https://github.com/sazabi/cve.now]	No terms of service or commercial-use rule is published; CC BY 4.0 on summaries allows commercial reuse with credit.
CORS / browser useAccess-Control-Allow-Origin: . The README says /feed and /api/v1/ send ACAO *; our GET and MCP POST returned it.	Yes⁴ [Verified 2026-10-28 by test request to https://github.com/sazabi/cve.now]	Access-Control-Allow-Origin: . The README says /feed and /api/v1/ send ACAO *; our GET and MCP POST returned it.
Freshness / coverageCVE List, GitHub, OSV and KEV every few minutes, NVD and EUVD every ten; count from /api/stats on 2026-10-28.	Minutes; 397,140 entries⁶ [Verified 2026-10-28 by test request to https://cve.now/api/stats]	CVE List, GitHub, OSV and KEV every few minutes, NVD and EUVD every ten; count from /api/stats on 2026-10-28.

What it does

cve.now is a free vulnerability feed run by Sazabi. It pulls the CVE List, the GitHub Advisory Database, OSV, NVD, ENISA EUVD, the CISA Known Exploited Vulnerabilities catalog, FIRST EPSS scores and Exploit-DB, and folds them into one record per vulnerability, with aliases such as the matching GHSA id. Each record keeps the CVSS score and vector, whether the bug is in KEV, and the affected and fixed versions, next to a short plain-English summary that a model wrote and that names the model. For agents, the useful part is the hosted MCP server with three tools: get_vulnerability looks up one CVE, GHSA or OSV id, list_feed returns the newest items for a filter, technology or package ecosystem, and check_stack takes a lockfile or a package list and returns the known advisories for those exact versions. The same data is available as plain JSON over HTTPS and as RSS, Atom or JSON Feed. The service launched on October 22, 2026, and its public GitHub repository holds docs and issues only; the code is private.

Working exampleRun 2026-10-28

ShellRequest

curl -sSL "https://cve.now/api/v1/entries/CVE-2021-44228"

Response (trimmed)

HTTP responseResponse

HTTP 302 -> 200 (run 2026-10-28; redirects to a static JSON copy on a CDN)
{"id":"CVE-2021-44228",
 "aliases":["GHSA-JFH8-C2JP-5V3Q"],
 "title":"Apache Log4j2 JNDI features do not protect against attacker controlled LDAP and other JNDI related endpoints",
 "summary":{"text":"Apache Log4j2, a widely used Java logging library, is affected, ...",
            "model":"anthropic/claude-haiku-5.5","promptVersion":"v1"},
 "severity":{"cvss":10,"vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H","source":"nvd"},
 "exploitation":{"kev":{"dateAdded":"2021-12-10","dueDate":"2021-12-24"},
                 "epssReached":{"date":"2026-10-22"},"exploitDb":null},
 "affected":[...], "references":[...], "sources":[...]}

Agent setup

Prompt to give your agent

Prompt

Add the cve.now MCP server (remote Streamable HTTP at https://mcp.cve.now/mcp, no authentication). Run check_stack on this project's lockfile. For every advisory it returns, give me the package, the installed version, the CVE id, the CVSS score, whether it is in CISA KEV, and the lowest fixed version. Sort by KEV first, then CVSS. Do not upgrade anything yet.

Agent / MCP-style tool snippet

Shell

# Claude Code
claude mcp add --transport http cve-now https://mcp.cve.now/mcp

JSON

# Cursor: .cursor/mcp.json
{
  "mcpServers": {
    "cve-now": { "url": "https://mcp.cve.now/mcp" }
  }
}

# Tools: get_vulnerability(id), list_feed(name|filter|tech|ecosystem, limit<=200),
#        check_stack(lockfile_content | packages[{ecosystem,name,version}] | stack_id)

Fit and safety

Good for

Coding agents that audit a lockfile before a release
Looking up a CVE id mentioned in a ticket or chat
Watching new KEV or critical entries for one framework
Feeding a Slack channel with a stack-specific RSS feed

Watch out

Do not treat an empty check_stack result as a clean bill of health. The tool itself says no matches means none found, not a guarantee; it misses advisories without version data and unsupported ecosystems.
Do not paste a private lockfile if package names are sensitive. Uploaded lockfiles are deleted after parsing, but the package names and versions are kept to build the stack feed, and anyone with a /stack?s= link can see the matches.
Do not quote the summaries as authoritative. A model writes them; take severity, KEV status and affected versions from the structured fields or the original description.
Do not build a compliance process that depends on it alone. It is a new, free service with no published SLA, rate limit or terms page; keep OSV or NVD as a fallback.

Alternatives

OSV API	Google’s free, keyless open-source vulnerability API, one of cve.now’s own sources, with a batch query for many packages.
NVD CVE API 2.0	The U.S. government’s CVE database with CVSS data; works without a key at lower limits, with a free key for more.
GitLab MCP server	When the agent also needs to open issues or merge requests for the fixes it finds.
Official MCP Registry API	To find other security MCP servers and compare them.

FAQ

Does cve.now need an API key?

No. The JSON API, the feeds and the MCP server at https://mcp.cve.now/mcp all work without a login or key.

How do I check my dependencies for CVEs with Claude?

Add the cve.now MCP server and ask Claude to run check_stack on your lockfile. It accepts package-lock.json, pnpm, yarn, bun, poetry, requirements.txt, Cargo.lock, go.sum, Gemfile.lock, mix.lock and flake.lock.

Where does cve.now get its data?

From the CVE List, GitHub Advisory Database, OSV, NVD, ENISA EUVD, CISA KEV, FIRST EPSS and Exploit-DB, merged into one entry per vulnerability.

Compare

Put cve.now API + MCP side by side with other free agent tools on keys, limits, MCP support, licensing and CORS. Open the comparison grid →

SourcesFootnotes

1cve.nowVerified 2026-10-28
2cve.now/llms.txtTested 2026-10-28
3cve.now/#faqVerified 2026-10-28
4github.com/sazabi/cve.nowVerified 2026-10-28
5github.com/sazabi/cve.now#sources-and-licensingVerified 2026-10-28
6cve.now/api/statsVerified 2026-10-28