Server has 12 tools with basic descriptions and parameter schemas. All tools follow verb_noun naming (ipf_get_*, ipf_set_*). However, descriptions are generic and lack LLM-optimized guidance on WHEN to use each tool vs. similar ones. Parameter descriptions are minimal (e.g., 'Optional dict of filters' lacks format/constraint details). No output schemas documented. Error handling returns generic error objects without recovery guidance. Most tools are READ_ONLY inventory queries with repetitive structure, suggesting composition issues. Baseline: avg tool description ~60 chars (well below 194-char production baseline); params lack type constraints (enums, ranges, patterns).

• Expand tool descriptions to 100 - 200 characters. Include WHEN to use each tool: 'Get device inventory (routers, switches, firewalls). Use this to find devices by hostname, vendor, or platform. Returns device ID, name, vendor, model, and IP address.'
• Document output schemas for all tools. Specify returned fields, types, and structure. Example: 'Returns: {"devices": [{"id": string, "hostname": string, "vendor": string, "model": string, "ip_address": string}], "total_count": integer}'
• Add format constraints to filter and columns parameters. Example: 'filters: dict with keys like hostname, vendor, model, site_id (see ipf_get_available_columns for valid column names). columns: list of column names to return; omit for all columns.'
• Implement error classification. Return structured errors: {"error": "not_found", "message": "Snapshot ID 'xyz' not found", "recovery": "Call ipf_get_snapshots() to list available snapshot IDs"}
• Consider consolidating the 8 similar inventory query tools into a single 'ipf_query_inventory' tool with a 'table_type' enum (devices|interfaces|sites|vendors|platforms|routing|vlans|neighbors) to reduce LLM confusion and improve composability.

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

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