Server has 6 tools with basic schemas and descriptions, but significant gaps prevent production use. All tools have descriptions (10-50 chars, below the 50-200 char baseline for LLM optimization), and input schemas are present but minimal. Parameters lack detailed constraints, enums, and format guidance. No output schemas documented. Error handling is generic ('Error: {str(e)}') with no recovery guidance. Tool names follow verb_noun convention (good), but descriptions are too terse to guide LLM selection. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels. Missing idempotency guarantees and pagination for list tools.

• Expand all tool descriptions to 50-200 characters. Example: 'list_leads' → 'Retrieve leads from a pipeline or stage with optional filtering. Returns lead ID, name, price, and status. Use to discover available leads before calling get_lead for details.'
• Document output schemas for all tools. Specify which fields are returned (lead_id, name, price, pipeline_id, stage_id, tags, created_at, etc.) so LLMs can chain calls and extract required IDs.
• Add tool annotations: readOnlyHint=true for list_* and get_*; destructiveHint=true for create_* (since they modify state); idempotentHint=true if create_lead with same name is safe to retry.
• Implement structured error responses: categorize as retryable (rate limit, timeout) vs user-fixable (invalid pipeline_id, missing required field) vs fatal. Return 'Invalid pipeline_id: 999. Available pipelines: [1, 2, 3]' instead of generic 'Error'.
• Add pagination to list_leads and list_contacts: accept 'offset' and 'limit' (default 50, max 100), return 'total_count' and 'has_more' so LLMs can iterate safely without context explosion.

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

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