barnacle-search exposes 8 well-named tools with clear verb-noun patterns (set_project_path, build_deep_index, search_code, semantic_search). All tools have descriptions (avg ~100 chars, within baseline 34 - 392). Input schemas are present and typed. However, parameter descriptions are sparse or missing context; output schemas are not documented; error handling lacks recovery guidance; no tool annotations (readOnlyHint/destructiveHint). The server is functional but lacks LLM-optimization details needed for confident tool selection and error recovery.

• Add output schemas to all tools. Document return types as JSON Schema (e.g., get_file_summary returns {language: string, line_count: int, imports: [string], exports: [string], symbols: [{name: string, type: string, line: int}]}). This enables LLMs to plan downstream calls.
• Annotate tools with readOnlyHint and destructiveHint. Mark set_project_path, build_deep_index as destructive; all others as readOnly. This guides LLM retry logic and safety reasoning.
• Expand parameter descriptions with constraints. E.g., 'pattern: fnmatch-style glob (e.g., *.py, src/**/*.ts), max 256 chars, no absolute paths' and 'limit: max results to return (1 - 100, default 10)'.
• Add error recovery guidance. E.g., 'If file not found, call find_files() first to locate it' or 'If project not set, call set_project_path() with the project root directory'.
• Document pagination for semantic_search and search_code. Add offset/cursor and total_count to responses. State in descriptions: 'Returns up to {limit} results. Use offset to fetch additional pages.'

Inferred effective spec: <=2025-11-25.

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