# vuln.mlab.sh > Free CVE search engine and dependency scanner. One record per CVE merged from NVD, the CVE List (CVE.org), OSV, CISA KEV, ENISA EU KEV and EUVD, FIRST EPSS, CISA Vulnrichment (SSVC) and national CERT advisories. No login and no API key for reads. Every CVE page has a Markdown copy: append `.md` to its URL. JSON endpoints answer with `Cache-Control` headers; respect them when polling. ## Look up CVEs - [One CVE as Markdown](https://vuln.mlab.sh/cve/CVE-2024-3094.md): replace the id. Fields: status, dates, CVSS 3.x and 4.0, EPSS, CISA KEV, EU KEV, EUVD id, SSVC, CWE, affected CPEs, references. - [One CVE as JSON](https://vuln.mlab.sh/api/v1/cve/CVE-2024-3094): the full `CveItem` record. - [Latest CVEs as Markdown](https://vuln.mlab.sh/cve/latest.md): the 50 newest, one table row each. - [Search](https://vuln.mlab.sh/api/v1/cve?q=fortinet&severity=CRITICAL): `q`, `vendor`, `severity` (CRITICAL, HIGH, MEDIUM, LOW), `minSeverity`, `minCvss`, `kevOnly=1`, `cwe=CWE-79`, `cpe=cpe:2.3:...`, `dateStart` / `dateEnd` (YYYY-MM-DD), `page`, `limit` (max 100), `sort` (`recent`, `severity`). - [Latest as JSON](https://vuln.mlab.sh/api/v1/cve/latest) - [Archive by day and week](https://vuln.mlab.sh/archive.md): CVEs published per day and per ISO week (UTC), with critical, high, not scored yet and CISA KEV entries. [JSON](https://vuln.mlab.sh/api/v1/archive) - [Bulk dump](https://vuln.mlab.sh/api/v1/cve/dump?dateStart=2026-11-01&dateEnd=2026-11-07&minCvss=7): rate limited per IP. - [KEV catalogs for one CVE](https://vuln.mlab.sh/api/v1/cve/CVE-2024-3094/kev) - [Press sightings for one CVE](https://vuln.mlab.sh/api/v1/cve/CVE-2024-3094/sightings) - [Trending CVEs](https://vuln.mlab.sh/api/v1/cve/trending): mentioned by 2+ outlets in the last 7 days. - [Early warnings](https://vuln.mlab.sh/api/v1/early-warnings): serious vulnerabilities the press reported before they had a CVE ID (30 days), grouped by event, with a confidence (`confirmed`, `corroborated`, `single`), the timeline of articles and the CVEs once assigned. One: `?id=`. Page: https://vuln.mlab.sh/early-warnings, RSS: https://vuln.mlab.sh/early-warnings/rss - [Ghost CVEs](https://vuln.mlab.sh/api/v1/cve/ghost): CVE IDs cited in the press but not published at CVE.org yet. - [National CERT advisories](https://vuln.mlab.sh/api/v1/advisories) - [Data source health](https://vuln.mlab.sh/api/v1/sources): last sync per source. ## Polling - Every JSON answer, feed and Markdown page has an `ETag`: send it back as `If-None-Match` and get `304 Not Modified` when nothing changed. - [What changed since a time](https://vuln.mlab.sh/api/v1/cve/latest?since=2026-11-20T10:00:00Z): CVEs published or modified since `since` (RFC 3339 UTC or a date, 7 days back at most), 200 per `page`. The answer's `until` is the `since` of your next poll; fetch more pages while `more` is true. ## Rankings - [This week's worst CVEs](https://vuln.mlab.sh/rankings.md?window=7d): top 50 by the mlab vuln index (0-100). Windows: `7d`, `30d`, `365d`, `all`; past weeks with `&week=2026-W47`. - [Same as JSON](https://vuln.mlab.sh/api/v1/rankings?window=7d) - Labels on every scored CVE (JSON `exposure` and `attention`, Markdown rows): exposure is "Actively exploited", "Default setups", "Specific setups only" or "Unclear"; attention is "Needs attention" (with its reasons) or "Reference". Rules: https://vuln.mlab.sh/rankings/methodology#labels - [How the score works](https://vuln.mlab.sh/rankings/methodology): exploitation 40, exposure 25, severity 15, attention 15, momentum 5. Each CVE's own score is in its JSON (`vuln_index`) and Markdown. ## Search syntax Operators in `q` work everywhere `q` does (search page, `/api/v1/cve`, feeds): `vendor:fortinet` and `product:fortios` (NVD CPE names), `tech:nextjs` (see [technologies](https://vuln.mlab.sh/tech)), `ecosystem:npm` (OSV), `severity:critical` (add `+` for "and above"), `cvss:>=9`, `cwe:79`, `is:kev`, `year:2024`, `after:2026-01-01`, `before:2026-06-30`, `"exact phrase"`, `-excluded`. Example: [`vendor:fortinet is:kev severity:critical`](https://vuln.mlab.sh/api/v1/cve?q=vendor:fortinet%20is:kev%20severity:critical). A search with no result returns `did_you_mean` when a typo is likely. Facet counts (severity, year, KEV, top vendors) for any search: [`/api/v1/cve/facets?q=...`](https://vuln.mlab.sh/api/v1/cve/facets?q=ssl%20vpn). ## Technologies 37 curated technologies (Linux kernel, Windows, Chrome, FortiOS, PAN-OS, Kubernetes, Next.js, React, Django, Log4j...), listed at https://vuln.mlab.sh/tech. Each one is `?tech=` on search, the API and the feeds, comma-separated for several; `?ecosystem=npm` filters on an OSV ecosystem. Example: [Next.js and React as JSON Feed](https://vuln.mlab.sh/rss?tech=nextjs,react&format=json). ## Feeds Any search filter above also works on the feed. Up to 20 entries without a filter, 50 with one. - [RSS 2.0](https://vuln.mlab.sh/rss) - [Atom 1.0](https://vuln.mlab.sh/rss?format=atom) - [JSON Feed 1.1](https://vuln.mlab.sh/rss?format=json): each item carries a `_vuln` object (cve, cvss, severity, epss, kev). - [Example: KEV with CVSS 9+ as JSON Feed](https://vuln.mlab.sh/rss?kevOnly=1&minCvss=9&format=json) - Per CNA: `https://vuln.mlab.sh/cna//rss` (short names at https://vuln.mlab.sh/api/v1/cna). ## Check dependencies - `POST https://vuln.mlab.sh/api/v2/scan` with the raw lockfile as the body (`curl --data-binary @package-lock.json`), or `GET /api/v2/scan?url=`. Format is detected from the content; force it with `?format=` or name it with `?filename=`. - Supported: package-lock.json, yarn.lock, pnpm-lock.yaml, bun.lock, flake.lock (each input's commit is checked at osv.dev), Cargo.lock, poetry.lock, uv.lock, pdm.lock, Pipfile.lock, requirements.txt (`==` pins only), composer.lock, Gemfile.lock, go.sum, pom.xml, gradle.lockfile, packages.lock.json, pubspec.lock, mix.lock, mise.lock, CycloneDX (JSON, XML), SPDX (JSON, tag-value). - Anonymous scans: 512 packages, one every 10 seconds. With a free API token (`Authorization: Bearer ...`, from https://vuln.mlab.sh/me/tokens): 1024 packages. - Malicious packages (OpenSSF `MAL-` records, GitHub malware advisories) are flagged: `vulns[].malicious` and `results[].malicious` are true, and `malicious_packages` counts them. Remove those, do not upgrade them. - `POST https://vuln.mlab.sh/api/v2/query`: OSV-compatible lookup for one package version. - [One package](https://vuln.mlab.sh/package/npm/lodash): `/package//` lists every advisory with affected ranges, malicious releases and registry downloads. Ecosystems: npm, pypi, crates.io, go, maven, nuget, packagist, rubygems, hex, pub. - [Most exposed packages](https://vuln.mlab.sh/api/v1/packages): the most downloaded npm, PyPI and crates.io packages with new advisories in the last 30 days (page: https://vuln.mlab.sh/packages). - `POST https://vuln.mlab.sh/api/v2/enrich`: KEV, EPSS and threat actors for a list of CVE ids. ## Agents - Minimal MCP server, anonymous, no key: `https://vuln.mlab.sh/mcp` (Streamable HTTP). Two tools: `get_cve` and `search_cves` (search syntax included). 60 calls per 10 minutes per IP. - Full MCP server: `https://mlab.sh/mcp`, with lockfile / SBOM scans, threat actors, KEV and EPSS enrichment, IOC lookups and more. Every feature is free; it only needs a free mlab.sh account for the token. Setup: https://doc.mlab.sh/docs/mlab.sh/integrations/mcp - Claude Code plugin: `/plugin install mlab@mlab-sh`, then `/mlab:sbom-audit`. ## More - [Full API reference](https://vuln.mlab.sh/api) - [Source code of the MCP server](https://github.com/mlab-sh/mcp)