Model Context Protocol
Vulnerability intelligence, inside your AI coding agent.
One URL and one header gives Claude Code, Cursor, VS Code, Windsurf, Zed and every other MCP client 31 security tools and 33 expert workflows. No CLI install, no local binaries, nothing to run.
Learn more at vulnetix.com →Your agent already reads your code. Now it can check it.
An AI coding agent will happily add a dependency, bump a version, or tell you a CVE “looks low risk”, with no access to whether that package is being exploited this week, whether CISA has put a deadline on it, or whether a patched version even exists.
The Vulnetix MCP server closes that gap. Point your agent at one URL and it gains 31 tools over a vulnerability database aggregating 150+ sources, plus 33 expert workflows written by security engineers. Ask “is this CVE being exploited?” and the answer comes from CISA KEV, EPSS, exploit-maturity signals and real in-the-wild sightings, not from what the model remembers.
What makes it different from a wrapper
Responses are shaped for a model, not for a browser. A raw CVE record from the database is 2.6 MB, roughly 650,000 tokens, which no context window can hold. Every tool reduces its response server-side before you ever see it. The Log4Shell record comes back at 10.6 KB with every decision field intact: KEV due date, SSVC decision, EPSS score, exploitation maturity, affected packages, the functions to grep your codebase for.
It is stateless, so there is nothing to run. No local process, no daemon, no
container, no npx shim. Protocol revision 2026-07-28 removed sessions from MCP
entirely, so the server holds nothing between requests. You add a URL to a config
file and you are done.
Counts are never hidden behind a plan. Where a plan gates the detail, meaning the individual exploit records and the raw IOC list, you still see the counts on every tier. The tool says plainly that the detail is withheld, so an empty list never silently means “nothing found”.
The AI layer can never break a tool. Five tools add a model-written analysis on top of the database result. If inference is unavailable, tier-gated, or fails, you still get the data and a note explaining why nothing was added.
What it is not
It does not scan your repository. A Cloudflare Worker has no filesystem, so anything that needs to read your files (SAST, secret scanning, SBOM generation, applying a fix) ships as a skill prompt that drives your own agent and the Vulnetix CLI locally. Sixteen of the 33 prompts work that way, and each one says so at the top.
No code ever leaves your machine through this server.