Four tools with complete input schemas and descriptions. lcm_grep and lcm_describe have detailed, context-rich descriptions (200+ chars) that guide LLM selection. Parameters are well-typed with enums and constraints. However, output schemas are not documented, LLMs cannot see what fields to expect from responses. lcm_expand_query is a stub that returns guidance rather than data, reducing its utility. No tool annotations (readOnlyHint, etc.) despite all being read-only operations. Error handling is present but minimal.

• Document output schemas for lcm_grep (return format: array of {id, snippet, score, type}?), lcm_describe (return format: {id, content, lineage?, source_messages?}?), and lcm_stats (return format: {total_conversations, total_messages, total_summaries, by_prefix: {...}}?). Include these in tool descriptions or as a separate schema reference.
• Add readOnlyHint=true to all four tool definitions. This signals to agents that these tools are safe to call without side effects and enables caching/retry optimizations.
• Remove or fully implement lcm_expand_query. If v2 is not ready, delete the stub tool. If you must keep it, make it a resource or prompt instead of a tool, so agents do not attempt to call it.
• Enhance error handling in lcm_grep and lcm_stats. Return structured errors with actionable guidance: 'Invalid FTS5 syntax in pattern. Example: "foo OR bar". See https://www.sqlite.org/fts5.html for syntax rules.'
• Add performance notes to parameter descriptions. E.g., 'limit: Maximum results to return (1-200, default 50). Larger limits may slow queries on large databases; use pagination for bulk exports.'