Autonomous red team tool for MCP servers. Finds exploitable vulnerabilities before attackers do.
This is a deliberately vulnerable mock server for red-team testing, not a production MCP tool. All 5 tools have minimal descriptions (10-20 chars), no parameter descriptions beyond the parameter name itself, and schemas that lack depth. Tool names are action-oriented (execute_query, read_file, execute_command, http_request, search_files) but descriptions are bare. No error handling guidance, no output schemas documented, no security constraints. The mock server intentionally exposes SQL injection, path traversal, command injection, and SSRF vulnerabilities, appropriate for a red-team tool but disqualifying for production use.
Execute a shell command
Execute a SQL query against the database
Make an HTTP request to a URL
Read a file from the filesystem
Search for files matching a glob pattern
All tool descriptions are under 20 characters and provide no context for LLM selection. Examples: 'Execute a SQL query against the database' (41 chars, acceptable) but parameter descriptions are missing entirely.
No parameter descriptions provided. Each tool's inputSchema lists properties but omits 'description' fields for parameters. LLMs cannot infer parameter meaning from names alone (e.g., 'query' could be SQL, search query, or GraphQL).
No output schemas documented. Tools return unstructured text responses (e.g., 'Contents of {path}') without declaring what fields or structure the LLM should expect. Agents cannot plan downstream calls or extract structured data.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | <=2025-11-25 | v2 |
No error handling or recovery guidance. Tools accept any input without validation (SQL injection, path traversal, command injection, SSRF all possible). Errors are not categorized as retryable, user-fixable, or fatal.
No security constraints or permission gates. Tools expose destructive operations (execute_command, execute_query) without permission checks, scope declarations, or audit trails. Credentials and secrets are not injected server-side.