Loading ToolBench
MCP server that provides access to Sinequa documentation search via neural search capabilities
Single tool with critical definition gaps. Tool name 'get_documentation' is appropriately verb-prefixed, but the description is vague (67 chars, below 100-char minimum for clarity). Input schema exists but lacks parameter descriptions, the 'query' parameter has only a typo-ridden description ('documentatino'). No output schema is documented. The tool performs browser automation with side effects (login, navigation) but lacks error handling guidance, timeout documentation, or recovery instructions. No tool annotations (readOnlyHint, etc.) despite being marked READ_ONLY. Overall definition quality is below production baseline.
Returns the html of the top documentation pages related to the query, as defined by Sinequa Neural Search
Parameter description contains typo ('documentatino' instead of 'documentation') and lacks format/constraint guidance. No mention of expected query length, format, or examples.
No output schema documented. Tool returns markdown text but LLM cannot predict structure, length, or fields. Violates pattern:response-shaper.
Tool description (67 chars) is below recommended minimum (100-200 chars). Does not explain WHEN to use it, what prerequisites exist (Sinequa credentials), or what happens on failure.
No error handling guidance. Tool can fail on login, navigation, parsing, or no results. Errors return raw Go errors with no recovery hints for the LLM.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
Tool performs stateful browser operations (login, navigation) but no timeout, retry, or idempotency documentation. Agents cannot safely retry on transient failures.