Loading ToolBench
Local, deterministic MCP server for auditing AI-writing patterns.
Two well-defined tools with complete input/output schemas, clear descriptions, and proper tool annotations. Both tools follow verb_noun naming (score_text, audit_text) and include context enums. Descriptions are substantive (150 - 180 chars) and explain when to use each tool. Input schemas use Zod with proper type constraints (minLength, maxLength, enum). Output schemas are fully documented with nested objects for statistics and highlights. Tool annotations correctly declare readOnlyHint=true and idempotentHint=true. Main gaps: no error handling guidance (e.g., what happens if text is malformed or detector fails), no parameter descriptions for 'context' enum values (what does 'technical' mode suppress?), and no recovery hints in descriptions.
Audit text locally with the deterministic Avoid AI Writing detector. Returns the score plus up to 100 flagged patterns and 100 highlighted sentence regions, with truncation counts. This is a heuristic writing audit, not proof of authorship; use score_text for a compact result.
Score text locally with the deterministic Avoid AI Writing detector. Returns a compact 0-100 pattern score, classification, confidence, and counts. This is a heuristic signal, not proof of authorship; use audit_text when individual findings are needed.
No error handling guidance in tool descriptions. If the detector fails or text is invalid, the LLM has no recovery path.
Context enum values lack descriptions. 'technical' mode is mentioned as suppressing code-adjacent patterns, but 'marketing' and 'personal' modes are unexplained.
No guidance on when to use score_text vs audit_text beyond 'compact' vs 'detailed'. LLMs may struggle to choose between them without clearer decision criteria.
Output schema includes 'label' field (e.g., 'Empty', 'Too short') but this field is not documented in the tool description. LLMs cannot infer its meaning or valid values.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 65 | 2025-06-18+ | v2 |