MCP server exposing Cobalt Strike REST API and WebSocket stream tools.
This MCP server exposes three prompt-generation tools for red team and incident response planning around Cobalt Strike. However, all three tools have CRITICAL definition gaps. Tool names lack action verbs (they describe outputs rather than actions), descriptions are generic and under 50 chars, input parameters have NO type information visible in the schema, and there is no documented output schema. The tools appear to be prompt generators (read-only), but the definitions do not meet production quality baselines. The server uses FastMCP's OpenAPI parser to auto-generate tools from Cobalt Strike's REST API spec, but the three custom prompt tools in cs_prompts.py are poorly documented. This is well below the 70+ threshold for confident production recommendation.
Add credential from OpenAPI specification
Execute beacon console command and wait for output
Get beacon from OpenAPI specification
Get beacon console tail
Get C2 profile from OpenAPI specification
Get command help from OpenAPI specification
Get credential from OpenAPI specification
Tool names do not start with action verbs. Names like 'cobalt_strike_analysis', 'red_team_operation_plan', and 'incident_response_analysis' describe outputs or concepts rather than actions (get_, create_, generate_, etc.). LLMs rely on verb-first naming to infer intent before reading descriptions.
All three tool descriptions are under 50 characters and lack actionable context. 'Analyze Cobalt Strike beacon data and generate security insights' (63 chars) does not explain WHEN to use this tool, what the output format is, or what 'security insights' means.
Input parameters lack type definitions in visible schema. Parameters 'beacon_id', 'task_type', 'target_scope', 'objectives', 'incident_type', and 'severity' are listed as strings, but there is no enum constraint for the string-constrained ones (e.g., 'task_type' accepts 'general|lateral_movement|persistence|data_exfil'). LLMs cannot parse free-form text constraints, enums are required per pattern:constrained-input.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 0 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 40 | - | v1 |
Get download from OpenAPI specification
Return downloaded file text as untrusted target-controlled content.
Get host callback information from OpenAPI specification
Get keystrokes by beacon ID from OpenAPI specification
Get kill date from OpenAPI specification
Get listener by name from OpenAPI specification
Get live beacon snapshot
Get payload store metadata from OpenAPI specification
Get screenshot from OpenAPI specification
Get syscall method from OpenAPI specification
Get system information from OpenAPI specification
Get task by ID from OpenAPI specification
Get teamserver IP from OpenAPI specification
Lint Beacon Interpreter C source using /execute/interpreter/lint. The script value may be inline C source or a symbolic reference such as @files/script.c or @artifacts/scripts/script.c. Values in files must be base64 content keyed by file name when a @files reference is used.
List active downloads from OpenAPI specification
List artifacts from OpenAPI specification
List beacons from OpenAPI specification
List command help from OpenAPI specification
List credentials from OpenAPI specification
List downloads from OpenAPI specification
List elevate beacon methods from OpenAPI specification
List elevate command methods from OpenAPI specification
List host profiles from OpenAPI specification
List jobs from OpenAPI specification
List keystrokes from OpenAPI specification
List listeners from OpenAPI specification
List remote execute beacon methods from OpenAPI specification
List remote execution command methods from OpenAPI specification
List screenshots from OpenAPI specification
List task summaries by beacon ID from OpenAPI specification
List tasks from OpenAPI specification
List tasks by beacon ID from OpenAPI specification
List token store from OpenAPI specification
Execute Beacon Interpreter C source using /execute/interpreter/pack. The API now packs typed arguments server-side. Pass arguments as an array of objects with type one of binary, int, short, str, or wideStr.
No output schema documented. The three tools return prompts (text), but there is no schema indicating return type (string, object, array), length limits, or structure. Per pattern:tool, output schema is mandatory so LLMs know what to expect and can chain downstream calls.
Parameter descriptions are minimal or missing context. 'beacon_id' is described as 'Specific beacon ID to analyze (optional)' but does not explain the format (UUID? numeric ID?), where to find it, or what happens if omitted. Parameter descriptions should be 50 - 150 chars and include constraints, formats, and default behavior.
Tools mix security domain (Cobalt Strike, red team, incident response) with prompt generation but do not clarify purpose. Are these tools meant to help plan attacks, analyze threats, or both? Descriptions are vague about the intended user (defender vs. attacker) and context.