Server has 4 tools with complete schemas and basic descriptions. Naming follows verb_noun convention (search_*, get_*, create_*). Schemas use Zod validation with proper types. However, descriptions are generic (34-72 chars, below 194-char baseline), lack WHEN/WHY context, and parameter descriptions are minimal. No output schema documentation. Error handling is basic (generic 400 errors). Missing idempotency guidance for create_sales_order. Tool composition is reasonable but descriptions don't guide LLM selection between similar tools.

• Expand tool descriptions to 150-250 chars. Include WHAT (action), WHEN (use case), and WHAT IT RETURNS. Example: 'Search invoices by customer name, state (draft/posted/paid), or limit. Returns invoice_id, amount, state, and due_date. Use this to find unpaid invoices before calling get_customer_balance.'
• Document output schemas for all tools. Specify returned fields, types, and whether results are paginated. Example: 'Returns array of {invoice_id: number, name: string, amount_total: number, state: enum, partner_id: number}. Limited to 10 results by default.'
• Add recovery guidance to error responses. Instead of 'Invalid partner_id', return 'Partner not found. Call search_customers(query="<name>") to find the correct partner_id.'
• For create_sales_order, add idempotency note: 'Idempotent: calling twice with same partner_id and product_lines creates one order. Retries are safe.' Also document required fields and validation rules (qty > 0, price_unit > 0).
• Add parameter dependency hints. E.g., 'If you only have customer name, call search_customers first to get partner_id.' This prevents wasted calls and guides agent planning.