MCP server for running CodeQL analysis on code repositories
Critical gaps across all dimensions. The 'analyze' tool has a basic schema but lacks parameter descriptions entirely, LLMs cannot infer what 'query', 'language', or 'path' mean or how to format them. Tool descriptions are minimal (10-20 chars) and do not explain WHEN to use the tool, WHAT it returns, or any prerequisites. The 'health_check' tool is trivial and should not be exposed as a callable tool. No output schema documentation. Error handling returns raw exceptions without recovery guidance. No input validation or constraint documentation (e.g., supported languages, query format). The code shows subprocess.run() with unsanitized user input, critical command injection vulnerability. No pagination, no structured output shaping, no idempotency guarantees.
Run CodeQL analysis on code
Health check endpoint
Parameter descriptions completely missing. 'query', 'language', and 'path' have no descriptions, LLMs cannot infer what format, constraints, or values are valid. Rubric requires every parameter to have a description explaining what it controls.
Tool descriptions are under 20 characters and provide no context. 'Run CodeQL analysis on code' does not explain WHEN to use it, what languages are supported, what the output contains, or how to interpret results. Rubric baseline: 194 chars average, 34-392 char range for A+ tools.
Command injection vulnerability. Code uses subprocess.run() with unsanitized user input (query.language, query.path) directly in shell command. An LLM could pass 'language; rm -rf /' and execute arbitrary commands. Requires input sanitization and validation.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 30 | 2026-07-28+ | v2 |
No output schema documented. The 'analyze' tool returns CodeQLResult but LLMs have no visibility into the structure of 'results' field (is it a list of dicts? what keys?). Rubric requires 100% of A+ tools to have documented return types.
health_check should not be a callable tool. It has no input schema (empty {}), trivial output, and no business value to an agent. Exposes unnecessary surface area and wastes LLM reasoning cycles. Remove or hide from tool list.