Loading ToolBench
Enterprise-grade security gateway for Model Context Protocol servers with authentication, policy-based access control, tool filtering, and audit logging for AI agents and LLMs
Only 1 tool is visible in source code (list-tools). The tool has a basic description and minimal input schema, but lacks comprehensive parameter descriptions, output schema documentation, and proper error handling guidance. The server appears to be primarily a gateway/proxy architecture rather than a tool-focused MCP server. Limited evidence of adherence to agentic tool patterns from the 54 pattern library.
List available tools from an MCP server, with policy-based filtering applied
Tool description is too brief (< 50 chars) and lacks context for LLM decision-making. Missing: when to use this tool, what it returns, relationship to other discovery tools.
Input schema lacks parameter descriptions. The 'mcp_server' parameter has no description explaining what value to pass (server name? URL? identifier?), violating the requirement that every parameter has a description.
Output schema is not documented. The tool description does not specify what the response structure is (e.g., array of tool objects? fields in each tool? pagination support?).
No error handling guidance. Missing: what happens if mcp_server is not found or unreachable? What should the LLM do next? No recovery hints provided.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 24 | - | v1 |
Policy-based filtering is mentioned in the description but not explained. The LLM cannot know what policies are applied, what tools might be filtered out, or how the filtering works.
Single tool exposed despite claiming to be an 'enterprise-grade' gateway with authentication, policies, audit logging. No tools for policy management, server configuration, user/role management, or audit access.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) visible in schema. list-tools is clearly read-only and idempotent but not declared.