Loading ToolBench
A policy-gated MCP server for version-controlled operational runbooks.
Three tools with clear, action-oriented names (list_, inspect_, execute_) and documented risk levels. Descriptions are present and explain state-changing behavior. Input schemas are visible with type definitions. However, parameter descriptions are minimal or absent in the schema (variables param lacks detail on expected structure), and output schemas are not documented in the source. Error handling guidance is limited. The approval token mechanism is sophisticated but underdocumented for LLM consumption.
Execute only commands declared in a version-controlled runbook. Requires the global execution gate; mutating plans also require an out-of-band approval token.
Validate variables and return the exact commands, plan hash, risk, and approval requirement without executing.
List version-controlled runbooks and their risk levels. Does not execute anything.
Output schemas not documented. LLMs cannot infer what list_runbooks, inspect_runbook, or execute_runbook return. Missing return type documentation violates pattern:tool and pattern:response-shaper.
Parameter 'variables' (object with additionalProperties) lacks description of expected structure, constraints, or examples. LLMs cannot determine valid variable names or formats without explicit guidance.
approvalToken parameter description is minimal ('Optional approval token for mutating runbooks'). Does not explain token format, how to obtain it, or consequences of omission. Violates pattern:tool-description completeness.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2025-06-18+ | v2 |
No error recovery guidance. If execute_runbook fails due to missing approval token or invalid variables, LLM receives no actionable next steps. Violates pattern:recovery-guide.
STDIO transport only. Server is not remotely accessible and cannot be used by hosted MCP clients. Hard cap at 50 for protocol readiness applies.