Loading ToolBench
Kubernetes incident investigation engine exposed as an MCP server over stdio. Provides read-only tools for investigating cluster incidents, replaying past investigations, listing and reading incident records, and previewing remediation actions.
KubeTective defines 5 tools with basic schemas and descriptions, but falls short of production quality. All tools have descriptions (10-60 chars, below the 50-200 char baseline for LLM optimization) and input schemas with type definitions. However, parameter descriptions are minimal or absent, output schemas are undocumented, and error handling guidance is missing. The tool names are reasonably clear (investigate, replay, list_incidents, read_incident, action_preview) but descriptions lack context on WHEN to use each tool or what downstream actions they enable. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) are present despite all being READ_ONLY. Schema completeness is moderate, parameters have types but lack constraints (enums, ranges, patterns) and detailed descriptions.
Preview remediation actions for a recorded incident without applying them
Run an investigation on a Kubernetes resource target
List all recorded incident IDs, newest first
Read a full incident record by ID
Replay a past investigation against a recorded incident
Descriptions are too short (10-60 chars vs. baseline 50-200 chars). 'Run an investigation on a Kubernetes resource target' lacks context on WHEN to call investigate vs. replay, what it returns, or how to interpret results. LLMs cannot determine tool selection from vague descriptions.
Parameter descriptions are missing or minimal. 'target' is described as 'Kubernetes resource reference (e.g., deployment/checkout, pod/my-pod)' but lacks format constraints, valid patterns, or error guidance. 'since' accepts '10m, 1h' but no min/max bounds or regex pattern is documented.
Output schemas are not documented. Callers cannot know what fields investigate() returns, whether it includes incident_id for chaining to read_incident(), or how to extract actionable data. This forces LLMs to guess and risks context loss between tool calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 43 | 2026-07-28+ | v2 |
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being READ_ONLY. Agents cannot determine which tools are safe to retry or whether they have side effects. This is a current MCP spec pattern (2026-07-28) that should be present.
No error handling guidance. If investigate() fails (e.g., resource not found, namespace invalid, timeout), the tool provides no recovery hints. Agents cannot self-correct or know whether to retry, ask the user, or try an alternative tool.