Server provides 8 read-only tools with consistent naming (verb_noun pattern: list_*, get_*, search_*). All tools have descriptions and input schemas with types. However, descriptions are generic and lack LLM-optimization guidance. Parameter descriptions are present but minimal (10-50 chars). Output schemas are not documented, responses are formatted as plain text summaries rather than structured JSON. No error handling guidance, no pagination metadata (total count, next_cursor), and no tool annotations (readOnlyHint). Field filtering via include_fields is a good pattern but underutilized. Missing: output schema documentation, error recovery guidance, and per-tool composition hints.

• Document output schemas as JSON objects with typed fields. E.g., list_subnets returns {subnets: [{id: int, cidr: string, usage: {used_pct: float}}], total: int, truncated: bool}. This enables LLM field extraction and chaining.
• Expand tool descriptions to 100-200 chars. Include WHEN to use (e.g., 'Use this to fetch a single subnet's details and optional addresses. Call list_subnets() first to discover subnet IDs.'), what it returns, and any prerequisites.
• Add error recovery hints to descriptions. E.g., 'If subnet_id is unknown, call list_subnets(section_id) first. Returns 404 if subnet does not exist.'
• Return pagination metadata in list_* responses: {items: [...], total: int, limit: int, offset: int, truncated: bool}. This lets LLMs know if more results exist and how to fetch them.
• Add tool annotations to all tools: readOnlyHint=true for all (since all are read-only). This signals to agents that these tools are safe to call without confirmation.