Today, we're launching cve.now, a free feed of the security vulnerabilities that actually need your attention. It merges the CVE List, CISA's list of exploited vulnerabilities, EPSS, and OSV into one entry per CVE, explains each one in plain English, and filters the firehose down to the ones that need action. You can get it as RSS, JSON, an API, or an MCP server for your coding agent. There's no login, no signup, and no API key. Try it at cve.now.

October 21 was a rough day for anyone running software in production.

• PHP2Shell: two bugs in core PHP that let an anonymous attacker run commands on a server with a single POST request. PHP runs up to 70% of the world's websites, and exploit attempts showed up before the bugs were disclosed.
• Node.js: a security release fixing ten vulnerabilities, including a bug in the Brotli decoder that allows remote code execution.
• LiteLLM: attackers hijacked the release pipeline and pushed a malicious version to PyPI and Docker that steals cloud and AI credentials.
• Railway: an attacker may have accessed customer environment variables - API keys, database credentials, and more - and Railway paused all deployments while everyone rotated their secrets.

Some of this you only hear about from the vendor. But most of it shows up first as a CVE or a security advisory. So part of the answer is obvious: subscribe to the sources that publish them, find out fast, and have someone check whether you're affected and patch it.

In 2026, that someone is increasingly your coding agent. And that's where things break down.

The raw vulnerability data is free and public. The problem is everything around it.

Your agent doesn't know what happened this week. Its training data is months old. When it searches the web, it mostly finds news articles and blog posts, not the actual records with affected versions and fixes.

The data is scattered across four places, in four shapes. The CVE List is a repository of hundreds of thousands of JSON files. CISA's KEV catalog (what's being exploited) and EPSS (the odds it will be) are separate files. OSV (which packages and versions are affected) is keyed by package. Answering “should I care about this one?” means pulling and joining all four, every single time.

Nothing tells your agent to look. Agents act when someone asks. A critical CVE can sit there for hours until someone on the team happens to see it on X.

And there's a lot to sift through. More than 48,000 CVEs were published last year - over 130 a day. Almost none of them matter to you. The trick is finding the few that do.

cve.now is a free feed of the security vulnerabilities that actually need your attention, built for agents and the people working with them.

1. It's current. It checks every source every five minutes, so your agent can know about a new CVE minutes after it's published.
2. It does the joining for you. Each CVE is one entry with all four sources merged in, and your agent gets it in a single MCP tool call.
3. It tells you when to look. The critical feed only gets a new item when a CVE needs action, so you can pipe it into Slack or use it to kick off your agent.

No login, no signup, no API key.

Every five minutes, cve.now pulls new and updated records from the CVE List, CISA KEV, EPSS, and OSV. It merges them into one entry per CVE, and only when records share an id (a CVE, GitHub, or OSV identifier). A CVE and its GitHub advisory never show up twice, and two different bugs never collapse into one.

The merge matters more than it sounds. Take CVE-2025-61624, a Fortinet bug from last week. Its severity score is 5.4, a “medium”. But CISA says attackers are actively exploiting it. An agent looking only at the CVE record would tell you to ignore it. On cve.now, it's in the critical feed.

Here's the kicker: the critical feed runs on a fixed rule, not an AI's judgment. An entry is critical if any of these is true:

• It's in CISA's KEV catalog. Someone is exploiting it.
• Its severity is 9.0 or higher, and it affects a widely used technology. A high score alone isn't enough, because many 9.0+ CVEs are in software almost nobody runs. We curate about 30 technologies today, from Next.js and Node.js to Kubernetes, nginx, PostgreSQL, and Fortinet.
• Its severity is 7.0 or higher, and it's likely to be exploited soon. That means an EPSS score of at least 0.1, which puts it in roughly the top 5% of all CVEs.

It's the same rule every time, so you're not paying an agent to re-triage 130 new CVEs a day, and you can see exactly why something made the cut. The rule is public, and we'll announce any change to it.

On top of the merged data, AI writes a short plain-English summary of each entry: what's affected and what an attacker can do. The summary is labeled as AI-written, and the original description is always right below it. Facts like severity, exploitation status, and affected versions always come from the source data, never from the model.

Everything is served as RSS, Atom, and JSON feeds, filterable by technology or package ecosystem (just PHP, or just npm), plus a JSON API and an MCP server.

In your coding agent. Paste this prompt into Claude Code, Codex, Cursor, or any other agent that supports MCP:

Add the cve.now MCP server to your MCP config. It's a remote server (Streamable HTTP) at https://mcp.cve.now/mcp. No auth or API key is needed. Name it "cve-now". Once it's connected, list the latest critical CVEs from it to confirm it works.

From then on, your agent can look up any CVE and check your dependencies against recent ones on its own.

In Slack. Install Slack's RSS app, then subscribe a channel:

/feed subscribe https://cve.now/feed?filter=critical

New critical CVEs show up in that channel as they land.

In any feed reader or script. Filters combine, so https://cve.now/feed?filter=critical&tech=php gets you critical PHP vulnerabilities only. It's RSS by default; add &format=atom or &format=json for the others. Every entry is also available as JSON:

curl https://cve.now/api/v1/entries/CVE-2025-61624

A feed like this is only useful if you know its limits, so here they are:

• History. cve.now has about the last 30 days of the CVE List. If it finds nothing for your dependencies, that doesn't mean you're safe from older vulnerabilities.
• Version matching works for package ecosystems that publish version ranges, like npm, PyPI, Go, Maven, and crates.io. Appliances and products outside our technology list show up in the full feed but aren't matched to your stack.
• Summaries are written for new entries as they arrive. Some older entries still show only the original description.
• No SBOM upload or OS packages yet.
• Incidents that never get a CVE won't show up here, or in any vulnerability feed. Railway's disclosure is one of them. More on that below.

The MCP server's tools only look things up. When your agent checks its dependencies, cve.now matches the packages against the catalog and keeps nothing. Vulnerability descriptions come back as plain data, which matters, because anyone who files a CVE writes its description.

Our summaries are licensed CC BY 4.0, and every entry lists its sources and their licenses.

Security is top of mind for everyone right now. After every big disclosure, our customers ask the same question: “Does this affect us?”

Sazabi already reads every log line and knows your codebase, so it sees what's actually happening in production. That makes it just as useful to security teams as it is to observability teams. On October 21, it flagged a customer's failing Railway deploys hours before Railway's security disclosure went out. That incident never got a CVE. No vulnerability feed would have caught it. The logs did.

cve.now is the other half: a clean, current catalog of what's been disclosed and what's being exploited. Sazabi reads the same feed to connect new CVEs to what it sees in your logs and your code.

We wanted the catalog itself to be useful to everyone, not just Sazabi customers, so we made it free. Huge shoutout to Alex Holovach for leading this one.

cve.now is just getting started. Next, we're adding more technologies, a longer history, and early warnings from vendor and CERT advisories, which often land hours before the CVE record exists.

In the meantime, we have three asks:

1. Try it. Add the MCP server to your agent or pipe the critical feed into Slack, and tell us when it gets something wrong at [email protected].
2. Tell us what to cover. Which technologies and vendors should we add next?
3. Share it. Know security or infrastructure folks who'd find this useful? Send them cve.now.

And if you want to know what's happening in your own systems, try Sazabi.

Sherwood Callaway CEO & [email protected]