cve-mcp-server
CVE MCP Server is a Python MCP server that lets Claude act as a vulnerability analyst. Instead of checking NVD, EPSS, CISA KEV, GitHub, VirusTotal, Shodan and GreyNoise by hand, you ask a question such as "Should we patch CVE-2024-3400?" and the server queries the relevant sources in parallel and computes a composite risk score. The README counts 27 tools across 21 data sources.
Tools are grouped into core vulnerability intelligence (`lookup_cve`, `search_cves`, `get_epss_score`, `check_kev_status`, `get_cvss_details`, `bulk_cve_lookup` for up to 20 CVEs and more), exploit and attack intelligence (`search_exploits`, `check_poc_availability`, MITRE ATT&CK and CAPEC mapping), plus network and threat-intelligence lookups. Architecture is a stdio server with an async HTTP client, rate limiting and a SQLite cache with an audit log. Many tools work without API keys; keys for NVD, GitHub and others are optional but recommended.
The README states what it does not do: no active scanning, no write operations apart from URLScan submissions, and no CVSS v4.0 scoring (only v3.1 is calculated). It sends CVE IDs, IPs, hashes and domains to the respective external APIs, blocks private and reserved IP ranges before any lookup, and loads keys from environment variables only. Known limits include 2,000 results per NVD query, 50 GreyNoise community queries a week and 4 VirusTotal requests a minute on the free tier. Licence is MIT, last push 2026-10-01. This review is based on the README and repository data; we have not run it ourselves.