Server has 4 tools with complete input schemas and basic descriptions, but significant quality gaps limit production readiness. All tools have descriptions (positive), but they are brief (avg 60 chars, below 194-char baseline) and lack actionable context. Schemas are present and properly typed via Zod, but descriptions lack parameter-level detail and error guidance. No tool annotations (readOnlyHint/destructiveHint) despite having destructive operations. Tool names follow verb_noun pattern but descriptions do not explain WHEN to use each tool or what happens on failure. bill_update explicitly states 'not supported', a red flag for agent reliability.

• Remove or implement bill_update. Current state ('not supported') breaks agent workflows. If unsupported, delete the tool and document workaround (delete + re-add) in bill_delete description.
• Add tool annotations: mark bill_query with readOnlyHint=true, bill_delete with destructiveHint=true, bill_add with idempotentHint=false (or true if API is idempotent). This enables agents to reason about retry safety.
• Expand tool descriptions to 100-200 chars. Example: 'bill_query: Query bills by natural language (e.g., "expenses in January", "transfers to savings"). Returns matching bills with amounts and dates.' Include WHEN to use it vs other tools.
• Document output schemas. Add a comment or JSDoc showing what bill_query, bill_add, bill_delete return (e.g., { bills: [{id, type, amount, date}], total, count }). Agents need to know what fields to extract.
• Enhance error messages with recovery hints. E.g., if fetch fails with 401, return 'Authentication failed. Ensure LEDGER_MATE_ACCESS_KEY environment variable is set.' If billId not found, return 'Bill not found. Call bill_query to list available bills.'