Both tools have reasonable names (verb_noun pattern) and descriptions (100-150 chars, within baseline 34-392 range). However, critical gaps undermine quality: input schemas are present but lack type information for several parameters (e.g., 'left', 'right' in calque_check are strings but no format/pattern constraints; 'seed_cmd' in calque_vocab_check is a string with no validation guidance). Parameter descriptions are minimal (10-30 chars) and fail to explain constraints, valid ranges, or dependencies. No output schemas are documented, LLMs cannot predict what these tools return or chain results to downstream calls. Error handling is absent from descriptions; no guidance on retryability, user-fixable vs fatal errors, or recovery paths. The tools appear to be CLI wrappers (--strict, --bootstrap flags) rather than true MCP tools with structured I/O, suggesting tool definitions may be inferred rather than explicitly registered in the MCP framework.

• Document complete output schemas for both tools. Specify field names, types, and meanings. Example: calque_check returns {suspects: [{file: string, line: number, reason: string}], total: number, registry_version: string}.
• Expand parameter descriptions to 50-100 chars with constraint details. E.g., 'min_score: Minimum suspicion score (0.0 - 1.0, default 0.5). Scores below this threshold are filtered out.'
• Add error handling guidance to tool descriptions. E.g., 'Exits with code 1 if --strict and violations found (retryable: no; user-fixable: yes, review registry or adjust thresholds).'
• Specify numeric bounds and enum values in schemas. Add minValue/maxValue for min_score, minItems/maxItems for exclude patterns, and enum for boolean flags.
• Verify tool registration in MCP framework. Ensure tools are explicitly registered via tool.register() or equivalent, not inferred from CLI parsing. Provide explicit JSON Schema for inputs and outputs.