Loading ToolBench
Simple, powerful governance for AI agents — policy enforcement, identity, sandboxing, and reliability in one package.
Agent Guard exposes 4 tools via HTTP, but critical definition gaps severely limit usability. Tool 1 (initialize) and Tool 4 (notifications/initialized) appear to be protocol-level concerns, not user-facing tools, their presence suggests confusion about MCP semantics. Tool 2 (tools/list) and Tool 3 (tools/call) are the only substantive tools. Tool 3 has a partial input schema visible (name, arguments, _meta), but Tool 2 lacks any visible schema. Descriptions are present but generic (10-50 chars), providing minimal LLM guidance. No output schemas documented. No parameter descriptions beyond the bare names. No error handling guidance. The server appears to be a governance/policy enforcement layer wrapping other tools, but the MCP interface itself is underdeveloped.
Initialize MCP protocol handshake and retrieve server capabilities
Notification that client has completed initialization
Execute a named tool with provided arguments and agent context
List all available MCP tools with their schemas and descriptions
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields without knowing response structure.
tools/list and initialize lack visible input schemas. tools/call schema is incomplete, no type constraints on 'arguments' object, no description of _meta structure or required fields.
Descriptions are generic and under 50 chars. 'Initialize MCP protocol handshake' and 'List all available MCP tools' provide minimal context for LLM tool selection. No guidance on when to call each tool or what to do with results.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 26 | <=2025-11-25 | v2 |
initialize and notifications/initialized appear to be protocol-level concerns, not user-facing tools. Their presence in the tool list suggests the server conflates MCP protocol mechanics with domain tools, confusing LLM tool selection.
No error handling guidance. No indication of which errors are retryable, user-fixable, or fatal. No recovery suggestions (e.g., 'If tool not found, call tools/list first').