Loading ToolBench
Kubernetes MCP server with progressive disclosure for discovering and executing code using discovered APIs
ProDisco exposes two tools with reasonably detailed descriptions and schemas, but several critical gaps prevent a higher score. Both tools have verb-starting names (search, run) and non-empty descriptions. However, the searchTools tool has an overly long, repetitive description (250+ chars, violating the guidance to keep descriptions concise). The runSandbox tool description is vague about what 'sandboxed environment' means and lacks error handling guidance. Parameter descriptions are present for most inputs, but the schema definitions reveal incomplete type coverage in some areas. Output schemas are not documented, and error recovery guidance is missing entirely. The server is STDIO-only, which caps protocol readiness at 50 regardless of other factors.
Execute code in a sandboxed environment with optional caching for discovered scripts. Supports custom timeouts and script caching.
Search for API members by name (methods/types/functions/scripts). Use a class/type/function/method name or keyword relevant to the libraries you configured. Searches indexed library APIs (prefers TypeScript typings; falls back to ESM JavaScript exports when typings are absent). No code execution.
Output schemas not documented for either tool. LLMs cannot plan downstream actions or extract return values without knowing what fields to expect.
No error handling guidance provided. Errors return raw messages without recovery hints or actionable next steps for the LLM (e.g., 'Try with a different query', 'Check sandbox resource limits').
searchTools description is overly verbose (250+ chars with repetition: 'Searches indexed library APIs' appears twice). Current wording wastes tokens.
runSandbox description is ambiguous: 'Execute code in a sandboxed environment' does not clarify what languages are supported (stated as TypeScript/JavaScript in the input schema, not the description), security model, or resource limits. LLM cannot determine when to call this vs. other execution tools.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 47 | - | v1 |
runSandbox has conflicting required parameters: both 'code' and 'cached' are marked optional, but the tool requires at least one. LLM will not know to provide one or the other without explicit mutual exclusion guidance in descriptions.
searchTools 'limit' parameter description lacks min/max context. Schema shows minimum=1, maximum=50, but the description says only 'Maximum number of results to return' without mentioning the minimum or default enforcement. LLM might try to request 500 results.
searchTools 'library' parameter enum values are not visible in the provided schema excerpt. The enum array is empty or the actual library names are not documented. Without knowing available libraries, LLMs cannot use this filter effectively.