ThreadShelf MCP defines 5 tools with complete JSON schemas and descriptions. Naming follows verb_noun convention (list_*, search, read_*). Descriptions are adequate (100-200 chars) but lack actionable context for LLM selection. All tools are READ_ONLY, reducing security risk. Schemas are well-formed with proper type definitions and enums. However, parameter descriptions are sparse (many lack guidance on format/constraints), output schemas are undocumented, and error handling is not visible in the tool definitions. The search tool is sophisticated (7 params with enums and constraints) but lacks pagination guidance despite potentially returning large result sets.

• Document output schemas for all tools. For search(), specify: {results: [{text, source_file, role, model, timestamp, collection}], total_count, next_cursor}. For read_thread(), specify: {turns: [{role, content, model, timestamp}]}.
• Add pagination to search() description: 'Returns up to n results. Use next_cursor (if present) to fetch additional results. Default n=15, max 50.'
• Enhance parameter descriptions with format rules. E.g., 'collection: Collection name (alphanumeric + underscore, case-sensitive) or "all" to search all collections. Defaults to "all".'
• Add error recovery guidance to tool descriptions. E.g., search: 'If no results found, try: (1) broaden query terms, (2) check collection name with list_collections(), (3) try keyword mode instead of semantic.'
• Specify what read_thread() returns on parse failure. E.g., 'Returns error with details: {error, reason, supported_formats}. If file format unsupported, call list_files() to verify file is indexed.'