Server has 5 tools with complete JSON Schema definitions and descriptions. Naming follows verb-noun patterns well (search-*, get-*, list-*). Descriptions are present but concise (50-80 chars typically). Schemas are well-formed with proper types, defaults, and constraints (e.g., maxResults, minScore ranges). However, descriptions lack LLM guidance on WHEN to use each tool and how they relate to each other. No output schemas documented. No error handling guidance visible. Tool composition is clear (each does one thing), but no documentation of chaining relationships or prerequisite calls.

• Expand tool descriptions to 100-150 chars explaining not just WHAT the tool does but WHEN to use it. Example: 'Search for annotations matching a natural-language query using semantic similarity. Use this when you don't know exact function names but have a description of what you're looking for. Alternative: use search-signature if you know the exact function signature.'
• Document output schemas inline or in a separate schema section. For each tool, specify the structure of the response, field types, and notably any IDs or references that can chain to other tools. Example: 'Returns an array of [{id: string, file: string, line: number, similarity: number}]. IDs can be passed to get-annotation().'
• Add error handling guidance to descriptions. Example: 'May return 'index_not_ready' if the annotation index is still building, retry after a few seconds. May return 'no_matches' if the query has no similar annotations, try broadening your search text or lowering minScore.'
• Document tool relationships. Example in search-similar description: 'Returns annotation IDs. Pass these to get-annotation() to retrieve full details. Or use search-signature() if you know the function name instead of the behavior.'
• Add optional pagination parameters to list-files and get-file-stats (limit, offset or cursor) if results can be large. Document the default limit and whether results are truncated. If not paginated, document the maximum expected result size.

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

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