Exposes the full NightShade/AdAgent toolset over the Model Context Protocol for autonomous Active Directory attack engagements
Single tool 'set_engagement' has a reasonable schema with 6 parameters and clear descriptions, but critical security and design flaws severely limit quality. The tool stores credentials server-side for injection into later calls, a dangerous pattern that violates secret-injection principles. Description is adequate (89 chars) but lacks guidance on when/why to call it. No error handling, no output schema documented, and the tool conflates authentication setup with engagement targeting (two concerns). Parameter descriptions are present but lack format constraints (e.g., what constitutes a valid FQDN, password complexity). The server exposes only one tool, limiting composition and agent flexibility.
Set the engagement target and credentials once before running any attack tool. NightShade injects these into later tool calls, so the host does not need to pass the password again. Provide password OR nt_hash.
Credentials stored server-side for injection into later tool calls violates secret-injection pattern. Passwords and NTLM hashes must never be retained across requests; use per-request authentication or OAuth-style tokens instead.
No output schema documented. Tool description does not specify what set_engagement returns, making it impossible for agents to verify success or extract confirmation data for downstream calls.
Tool combines two concerns: authentication (username/password/nt_hash) and engagement targeting (dc_ip/domain). Split into separate tools (authenticate_to_domain, set_engagement_target) so agents can compose them independently.
No error handling or recovery guidance. Description does not explain what happens if DC is unreachable, credentials are invalid, or domain does not exist. Agents have no path forward on failure.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Parameter descriptions lack format constraints. 'dc_ip' should specify IPv4 format; 'domain' should clarify FQDN format (e.g., 'corp.local'); 'nt_hash' should specify hex length (32 chars). LLMs cannot infer these from names alone.