Local MCP server for conversational CVE database queries
Three tools present with basic schemas and descriptions. All three are READ_ONLY and follow a consistent naming pattern (verb_noun). However, descriptions are minimal (under 100 chars each), parameter descriptions are sparse or missing, and output schemas are entirely undocumented. Tool names are clear but descriptions lack LLM guidance on when to use each tool versus alternatives. Parameters have basic type info but lack format constraints, ranges, and dependency documentation. No error handling guidance provided, agents do not know what to do if a CVE is not found or a search returns no results.
Get detailed information about a specific CVE by its ID. Returns severity, CVSS score, description, and references.
Get database stats (count, date range, last update)
Search CVEs by keyword in description
Output schemas entirely undocumented. Tools return JSON via TextContent but no schema definition provided for what fields agents should expect. E.g. get_cve_details returns parsed references, severity, cvss_score, none of these documented.
Parameter descriptions insufficient. 'cve_id' only says 'The CVE identifier (e.g., CVE-2024-0001)', no format constraint, range, or what happens on invalid input. 'keyword' lacks guidance on minimum length, special characters, or matching behavior. 'limit' has range 1-50 but no explanation why that range matters or what happens if exceeded.
No error handling guidance. If get_cve_details receives an invalid CVE ID, code returns {"error": "not found", "cve_id": ...} with no guidance to the agent on what to do next (try search_cves? ask user?). If search_cves finds zero results, agent sees {"count": 0, "results": []} with no message about constraints or suggestion to try broader terms.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 47 | - | v1 |
Tool descriptions too generic. 'Search CVEs by keyword in description' does not explain when to use this vs. get_cve_details; does not specify what fields are searched; does not clarify if partial matches work; does not mention performance characteristics (code comment notes FTS5 TODO). 'Get database stats' lacks context on what 'stats' means or why an agent would call it.
No input validation documented. Code normalizes CVE IDs (strip, uppercase, add prefix) but does not specify constraints in schema or parameter description. Agents cannot know if 'cve-2024-0001' (lowercase) is valid without trying. Limit parameter defaults to 10 but maxes at 50, no explanation of why or what happens if omitted.
Missing pagination contract. search_cves accepts 'limit' but no offset/cursor/page parameters. Results truncate descriptions at 300 chars without documenting this truncation in the tool schema or return contract. No 'total_count' or 'has_more' field to guide pagination logic.
No field naming consistency with chain-ability. get_cve_details returns references as an array, but there is no documented schema for what each reference object contains (URL? title? type?). search_cves returns 'cve_id', 'severity', 'cvss_score', 'description', 'published_date', but does not document whether agents can pass returned cve_id directly to get_cve_details without further lookup.