Loading ToolBench
Zero-trust runtime security and governance for AI agents with policy evaluation, security analysis, audit logging, and tool execution management.
The server exposes only one tool: 'echo'. The tool has a trivial description ('Returns received input') that is under 20 characters and provides no actionable context for LLM selection. The input schema declares type 'unknown' instead of a proper JSON Schema type (string, object, array, etc.), violating schema fundamentals. No output schema is documented. The tool lacks any error handling guidance, parameter constraints, or composition context. This is a minimal proof-of-concept, not a production-grade MCP server.
Returns received input
Input schema uses 'unknown' type instead of valid JSON Schema type (string, object, array, number, boolean). This violates JSON Schema spec and prevents LLMs from understanding what data structure to pass.
Tool description is 24 characters ('Returns received input'), below the 50-character minimum for LLM clarity. Does not explain WHEN to use this tool, WHAT it returns, or any prerequisites.
No output schema documented. LLMs cannot plan downstream tool calls or extract data without knowing the response structure.
Input parameter 'input' has no description. LLMs cannot infer what data types or formats are valid.
No error handling guidance. If the tool fails, LLMs have no recovery path or next-step instructions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 20 | 2026-07-28+ | v2 |