Starstruk Logo MarkStarstruk

v0.2.0

msitarzewski/agency-agentsv0.2.0Jun 22, 2026by mukul975

AI Summary

This release introduces a new `triage_cve` orchestrator for concurrent CVE risk assessment and implements significant security hardening measures. It adds support for new data sources, a new scoring algorithm, and infrastructure improvements.

Key Highlights

  • New `triage_cve` orchestrator with concurrent data fetching from NVD, EPSS, and CISA KEV.
  • Implemented comprehensive security hardening including SSRF defense and centralized secrets.
  • Added new data sources: VulnCheck NVD++, CIRCL hashlookup, and HIBP Pwned Passwords.
  • Infrastructure improvements with multi-stage Dockerfile and SQLite WAL support.

New Features

  • triage_cve orchestrator with quick, standard, and deep analysis modes.
  • CISA KEV hard override ensuring CRITICAL severity for known exploited vulnerabilities.
  • New MCP Resources: kev://catalog, epss://scores/{cve_id}, and manifest://tool-hash.
  • New MCP Prompts: patch_decision, compare_and_prioritize, and dependency_triage.
  • Streamable-HTTP transport support via MCP_TRANSPORT=http.

Full Release Notes

## CVE MCP Server v0.2.0

A major feature release. **Additive** — all existing tools remain registered; the new `triage_cve` orchestrator is the recommended entry point.

### ⭐ Orchestration
- **`triage_cve`** — one tool call fans out NVD + EPSS + CISA KEV (+ public PoC discovery) concurrently, computes the composite risk score, and returns a clean report. `depth` = `quick` / `standard` / `deep`; **`deep`** also emits an SSVC v2 gated decision (CISA Deployer model → `Act` / `Attend` / `Track*` / `Track`).
- Transparent **VulnCheck NVD++ fallback** used automatically when NIST NVD is unreachable or throttled.

### 🎯 Scoring
- **CISA KEV hard override** — a KEV-listed CVE is always **CRITICAL** (score clamped ≥ 76), regardless of CVSS/EPSS.
- **`scoring_version`** stamped into `triage_cve`, `calculate_risk_score`, and `health_check`.
- CVSS reframed as a **severity** signal (not exploitation-likelihood), per Allodi & Massacci 2014.

### 🔌 New data sources
- **VulnCheck NVD++** (NVD-schema fallback), **CIRCL hashlookup** (known-good allowlist), **HIBP Pwned Passwords** range API (keyless k-anonymity).

### 🧩 MCP enhancements
- **Resources**: `kev://catalog`, `epss://scores/{cve_id}`, `manifest://tool-hash` (SHA-256 over the registered tool surface).
- **Prompts**: `patch_decision`, `compare_and_prioritize`, `dependency_triage`.
- **Streamable-HTTP transport** via `MCP_TRANSPORT=http` (stdio remains the default).
- Expanded `health_check` (scoring version, manifest hash, configured-key status).

### 🔐 Security hardening
- Server **never** registers a sampling handler / never issues `sampling/createMessage` (Unit 42 MCP-sampling attack vector).
- **SSRF** defense: scheme allowlist + resolve-then-validate (rejects private/reserved IPs).
- Secrets centralized as Pydantic `SecretStr`.

### 🏗️ Infrastructure
- Multi-stage non-root **Dockerfile** + `.dockerignore`.
- SQLite **WAL** pragmas for concurrent reads; null-object rate limiter; fixed-interval deterministic retry; async fan-out helper; static Admiralty source-trust table.

**Verified:** ruff clean, 30/30 tests passing.

**Full Changelog:** https://github.com/mukul975/cve-mcp-server/compare/v0.1.0...v0.2.0


## What's Changed
* fix(7): set correct default values for cache and audit by @fergoid in https://github.com/mukul975/cve-mcp-server/pull/8

## New Contributors
* @fergoid made their first contribution in https://github.com/mukul975/cve-mcp-server/pull/8

**Full Changelog**: https://github.com/mukul975/cve-mcp-server/compare/v0.1.0...v0.2.0