A modular MCP providing AI-driven vulnerability management skills, including severity classification and automated insights.
VulnMCP demonstrates solid naming conventions (verb_noun pattern across all 19 tools) and comprehensive parameter descriptions. However, output schemas are not explicitly documented in the source code, and several tools lack clarity on return structure. Tool descriptions are generally adequate (100-300 chars) but some are overly technical without clear 'when to use' guidance. Error handling is minimal, no recovery guidance or actionable error messages visible. The server exposes API keys via environment variables (correct pattern) but lacks per-tool permission declarations and audit trail guidance.
Predict MITRE ATT&CK (Enterprise) techniques for a vulnerability. Uses CIRCL's fine-tuned RoBERTa-base multi-label model, trained on a curated gold set of CVE-to-technique mappings. Every technique in the model's vocabulary is scored independently (sigmoid); scores at or above 0.5 are the model's positive predictions, and the rest of the ranking shows how it orders the remaining candidates.
Classify a vulnerability description into CWE categories. Uses CIRCL's fine-tuned RoBERTa model to predict the most likely CWE (Common Weakness Enumeration) categories, mapped to their parent CWEs.
Classify the severity of a vulnerability based on its description. Uses CIRCL's fine-tuned transformer models: - English descriptions: RoBERTa-base model (low/medium/high/critical) - Chinese descriptions: MacBERT-base model (low/medium/high) - Russian descriptions: ruRoBERTa-large model (low/medium/high/critical) Language is auto-detected from the text unless explicitly specified.
Create a sighting for a vulnerability. Requires an API key on most Vulnerability Lookup instances. Set VULNMCP_API_KEY to authenticate automatically.
Get the full record of one CNA partner of the CVE Program. The short name is the CVE Program identifier for the partner — the same value that appears as assignerShortName in CVE records — so this tool can identify the authority behind a CVE returned by get_vulnerability. Use search_cna_partners to discover short names.
Output schemas not documented in source code. Tools return data via extract_summary() and normalize_payload() helpers, but the exact structure (field names, types, nesting) is not visible in tool definitions. LLMs cannot plan downstream calls without knowing what fields to expect.
No error handling or recovery guidance visible. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. E.g., 'API key required' or 'Vulnerability not found' have no suggested recovery steps.
No per-tool permission declarations. Tools like create_sighting (WRITE) and get_vulnerability (READ_ONLY) are marked with risk levels but lack explicit scope declarations (e.g., 'read:vulnerability', 'write:sighting'). Agents cannot be configured with least-privilege access.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2026-07-28+ | v2 |
Get a specific GNA entry by its numeric ID or exact short name. Exactly one of id or short_name must be provided.
Get vulnerabilities ranked by sighting activity. This helps quickly prioritize vulnerabilities with high real-world activity and distinguish potentially important issues from minor ones.
Fetch the 3 most recent vulnerabilities for a given CWE ID from Vulnerability Lookup.
Get detailed information about a vulnerability by its ID from Vulnerability Lookup. The base URL defaults to https://vulnerability.circl.lu and can be overridden with the VULNMCP_LOOKUP_URL environment variable.
Infer likely CPE identifiers from product keywords.
List GCVE references (vulnerability dataset sources and their GNA mappings). This includes KEV catalog entries with their Vulnerability-Lookup origin UUIDs, which can be used with the list_kev_entries tool's vulnerability_lookup_origin parameter to query a specific catalog.
List all Global Numbering Authorities (GNA) from the GCVE registry. Downloads and verifies the registry if not already cached locally.
List and filter Known Exploited Vulnerability (KEV) entries.
Search curated vulnerability bundles.
Search the CNA partners of the CVE Program. All filters are optional and combined with AND; with no filter, every partner is returned. Partners are CVE Numbering Authorities and related organizations as published on cve.org, mirrored at gcve.eu.
Search community comments related to vulnerabilities.
Search for GNA entries by name (case-insensitive substring match).
Search sightings to identify actively discussed or exploited vulnerabilities.
Search vulnerabilities on Vulnerability Lookup with filters. The base URL defaults to https://vulnerability.circl.lu and can be overridden with the VULNMCP_LOOKUP_URL environment variable.
Pagination parameters (per_page, page) present in many tools but no documentation of max limits or default behavior. LLMs may request 1000 results per page, causing timeouts or context window exhaustion.
Tool descriptions lack 'when to use' guidance. E.g., classify_severity, classify_cwe, and classify_attack_techniques all take vulnerability descriptions but do not explain when to call each vs. calling get_vulnerability first. Agents may waste calls on redundant classification.