CodeNavigator has 9 tools with basic schemas and descriptions, but falls short of production quality on multiple dimensions. All tools have input schemas (good), but descriptions are inconsistent, some are clear (analyze_codebase: 'Builds code graph...'), others generic (project_statistics: 'Get high-level project overview'). Parameter descriptions exist but are minimal (e.g., 'Name of the symbol to find definition for' is adequate but not LLM-optimized). No output schemas are documented anywhere. No error handling guidance is visible. No tool annotations (readOnlyHint, etc.). Tool naming follows verb_noun convention (find_definition, find_references) which is correct, but tool composition has issues: find_definition and find_references are distinct, which is good; however, find_callers and find_callees operate on 'function' parameter only, should they accept 'symbol' for generality? The analyze_codebase tool has an unusual requirement ('ALWAYS run this first') that is an anti-pattern, tool ordering should be inferred from context, not mandated in descriptions. Overall, the server is competent but mediocre, definitions are functional but lack the polish, error guidance, and schema documentation required for confident agent deployment.

• Document output schemas for all tools. For example, find_definition should return {symbol_name: string, file: string, line: number, type: string} with descriptions. Use examples to show LLMs what to expect.
• Rewrite descriptions as LLM-optimized prompts. Example for project_statistics: 'Returns project metrics: total files, lines of code, number of modules, cyclomatic complexity distribution, health grade (A - F). Call this to understand codebase scale before detailed analysis.'
• Unify parameter naming across tools. Use 'symbol' consistently for find_definition, find_references, find_callers, and find_callees instead of mixing 'symbol' and 'function'.
• Remove prescriptive language ('ALWAYS run this first') from analyze_codebase. Instead, add to description: 'Call this to build an initial code graph index. Results enable faster lookups by find_definition, find_references, etc.'
• Add error handling guidance to tool descriptions. Example: 'If symbol not found, returns empty array with message. Try search_symbols() to discover available definitions.' Use recovery-guide pattern.
• Add tool annotations to indicate read-only status. Use readOnlyHint: true in tool metadata so clients know these tools are safe to retry and cache.
• Document complexity_analysis threshold parameter: 'Minimum cyclomatic complexity threshold to report. Range 1 - 50. Default 10. Higher values show only most complex functions; lower values reveal all functions above threshold.'
• Add pagination support to tools that return lists (find_references, find_callers, dependency_analysis). Include limit, offset, and total_count in output schema to prevent context window overflow.

Inferred effective spec: 2026-07-28+.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry