Six tools with READ_ONLY operations, but critical gaps in schema visibility and parameter documentation. Tool names are action-oriented (search, callers, callees, imports, impact, raw), but input schemas are minimally documented. The 'search' tool has a reasonable description (hybrid semantic + BM25), but others lack depth. Parameter descriptions exist but are terse (e.g., 'Function name to find callers for'). No output schemas are documented in the source. Error handling and recovery guidance are absent. The 'raw' tool exposes raw Cypher queries, which is a security and usability risk without validation or guidance.

• Document output schemas for all six tools. Specify field names, types, and whether results are paginated. Example: search() returns {results: [{filename, location, code, score}], total_count, next_cursor}.
• Expand parameter descriptions to 50 - 150 characters each. Include format hints: 'function: string, fully-qualified name (e.g., module.ClassName.method_name) or simple name if unambiguous'.
• Add error recovery guidance to tool descriptions. E.g., 'If function not found, returns empty list. Try search() to discover available functions.'
• Implement input validation for raw tool. Reject queries that modify state (CREATE, DELETE, SET). Return actionable error: 'Raw Cypher queries must be read-only. Use MATCH and RETURN only.'
• Document pagination and result limits. Specify max results returned and whether tools support offset/limit parameters to prevent context window exhaustion.