Skip to content

Releases: MRFrazer25/Vulnerability-Exposure-MCP

Release list

v1.0.0

Choose a tag to compare

@MRFrazer25 MRFrazer25 released this 25 Oct 05:38

First public release.

Tools

  • investigate_target classifies a public IP, domain or URL, CVE ID, software version or distro package build (such as OpenSSH 8.9p1 Ubuntu-3ubuntu0.6), runs the baseline lookups, and returns an investigation ID, next steps and the full analyst guide.
  • Lookups: Shodan (or its free InternetDB without a key), VirusTotal, AbuseIPDB, the internet registries (RDAP), Spamhaus DROP, GreyNoise with a SANS ISC fallback, DNS, and subdomains from certificate transparency logs (crt.sh, with Cert Spotter as a fallback).
  • version_specific_vulnerability_check confirms a CVE only when NVD's affected-version ranges include the exact version, and adds GitHub and PyPA advisories (via OSV.dev) for packages NVD hasn't analysed yet. Results are ranked by CISA KEV, EPSS and CVSS, with each CVE's fix version and published exploits (Exploit-DB).
  • distro_patch_check uses Ubuntu's and Debian's own security data (via OSV.dev) to find CVEs a distro has already fixed by backport.
  • nvd_lookup, cve_search and cisa_kev_check, with Shodan CVEDB as a fallback when NVD can't be reached.
  • get_key_facts builds the report's key facts from the data in code, and lists which sources were and weren't checked.
  • check_report compares a drafted report with the investigation's data and flags what it doesn't support: CVE IDs no lookup returned (looked up in NVD to tell real ones from wrong ones), versions and percentages not in the data, unsupported claims, misfiled CVEs, misquoted fixes, upgrade versions named without a target version, and exploit claims the data doesn't back.
  • check_api_keys shows which keys are configured, never their values.

Security

  • API keys come from the OS credential store or the client's environment, and are never sent to a host other than the one they belong to.
  • Outbound requests use HTTPS only, follow redirects by hand, refuse private addresses, and cap response size.
  • Private, loopback, link-local, CGNAT and cloud-metadata addresses are refused, and IP lookups are limited to addresses the investigation has seen.
  • Third-party text has instruction-like content removed before the model sees it.

Clients

  • Works over stdio with Claude Desktop (including as a .mcpb bundle), Claude Code, Cursor, VS Code, Windsurf and other local MCP clients, and runs in Docker.

Install with uvx vulnerability-exposure-mcp or pip install vulnerability-exposure-mcp, or open the attached .mcpb file in Claude Desktop. See the README for setup.