Four read-only tools with clear, descriptive docstrings and proper parameter documentation. Tool names follow verb_noun convention (mm_accounts, mm_transactions, mm_portfolio, mm_categories). All tools have descriptions (194 - 250 chars, within baseline 34 - 392 range). Parameters are documented with context (e.g., 'account name exactly as shown by mm_accounts'). However, input schemas lack formal JSON Schema type declarations, parameters are documented in prose but not in structured schema. Output schemas are not documented. No error handling guidance for LLMs (e.g., 'if account not found, call mm_accounts first'). No pagination or result limits documented despite potential for large exports. Tool descriptions could better explain WHEN to use each tool vs. alternatives.

• Add formal JSON Schema to all tool inputs. For mm_transactions, declare from_date and to_date as 'type: string' with 'format: date' (ISO 8601). Mark from_date as required, to_date as optional. Add pattern or description: 'YYYY-MM-DD'.
• Document output schemas for each tool. E.g., mm_accounts returns list[dict] with fields: name (str), balance (float), currency (str), type (str), iban (str), uuid (str). mm_transactions returns dict with keys: transactions (list), account (str), date_range (dict).
• Add error recovery guidance to tool descriptions. E.g., mm_transactions: 'If account not found, call mm_accounts() first to list valid account names. Dates must be ISO 8601 (YYYY-MM-DD) and from_date <= to_date.'
• Clarify mm_transactions behavior when to_date is omitted. Formalize: 'If omitted, defaults to today's date (inclusive).' Add validation in code to reject from_date > to_date with a clear error message.
• Document result limits and pagination. E.g., 'mm_accounts returns all accounts (typically 5 - 50). mm_categories returns the full category tree (typically 20 - 200 nodes). No pagination available; if results are large, consider filtering by account type or date range.'