AI coding agents

Claude Code, told what it is about to install

Hooks, skills and MCP — the full surface, verified end to end.

Get a Free API Key

What one command wires

Wire Claude Code to Vulnetix: a dependency guard that answers at the moment the agent adds a package, 33 Agent Skills, and 31 MCP tools over a vulnerability database aggregating 150+ sources.

vulnetix agent install --agent claude-code

Where Claude Code reads skills

Agent Skills is an open standard, but hosts disagree about the directory. The installer writes the one Claude Code actually opens.

Add the MCP server

claude mcp add --transport http vulnetix https://mcp.vulnetix.com/mcp \
  --header "Authorization: ApiKey $VULNETIX_ORG:$VULNETIX_KEY"

True of Claude Code specifically

Install lodash at a version with known critical advisories

The guard refuses the tool call and gives the agent a reason it can act on. Claude Code stops and explains, in its own words, what it was told.

# wire every agent on this machine, once
$
 vulnetix agent install
Claude Code
  hooks    configured ~/.claude/settings.json
  skills   ~/.claude/skills
  mcp      https://mcp.vulnetix.com/mcp

# then ask the agent to add a dependency
>
 install lodash 4.17.20
→ npm i lodash@4.17.20
lodash@4.17.20 · npm
12 vulnerabilities · 6 critical, 6 high
- CVE-2021-23337  critical · CVSS 9.3 · EPSS 21.3%
- CVE-2025-13465  critical · CVSS 9.8 · EPSS 1.6%
- CVE-2026-4800   critical · CVSS 9.8 · EPSS 2.8%
- …and 9 more
Exploits: 324 known
Quick fix: bump to 4.17.23

STOPPED. Vulnetix gate block install — lodash@4.17.20
got 12 vuln (6 critical), 324 known exploit.

One stage of four