locidx provides 4 tools with complete input schemas and descriptions. Naming follows verb_noun convention (search, index, stats, roots). Descriptions are adequate (50-150 chars) but lack actionable context for LLM selection. Parameter descriptions are present but minimal. Output schemas are not documented, LLMs cannot predict response structure. Error handling is generic (tool failed: %s). No tool annotations (readOnlyHint/destructiveHint). The 'roots' tool exposes a deprecated MCP pattern. Overall: functional but below production baseline for agent-friendly tooling.

• Document output schemas for all tools. E.g., search returns {results: [{path, line_number, line_text}, ...], count: int}. Include this in tool descriptions or a separate schema field.
• Enhance descriptions with prerequisites and selection context. E.g., 'Search indexed files for a literal substring. Call index() first to populate the database. Returns matching lines with file path and line number. Use this instead of roots() when searching within a specific project.'
• Add tool annotations: mark 'search' and 'stats' with readOnlyHint=true; mark 'index' with destructiveHint=true (overwrites existing index) and idempotentHint=true (safe to retry).
• Replace 'roots' tool with a 'list_indexed_roots' parameter on search/stats, or document it as a discovery-only tool with clear guidance: 'Call this to discover available indexed projects before searching.'
• Improve error messages with actionable recovery steps. E.g., catch FileNotFoundError and return 'Directory not found: /path/to/root. Verify the path exists and is readable.' Catch empty query and return 'Query cannot be empty. Provide a literal substring to search for.'