MCP server for Sberbank — business accounts, balance, statements, payments, counterparties (Russia)
Server has 8 tools with consistent naming (verb_noun pattern) and tool annotations present. All tools have descriptions (10-200 chars, within baseline). Input schemas visible for 6/8 tools via Zod imports; 2 tools (list_counterparties, get_company_info) lack explicit inputSchema in registration, capping their schema scores. Parameter descriptions are present but terse (avg ~30 chars vs baseline 72). No output schema documentation. Error handling and recovery guidance absent. Composition is sound (single responsibility per tool), but descriptions lack WHEN/WHY context that LLMs need for tool selection.
Создание платёжного поручения. Денежная операция: идемпотентна по RqUID (повтор с тем же ключом не создаёт дубль).
Список счетов клиента в Сбербанке.
Баланс по счёту.
Сведения об организации-клиенте (реквизиты, ИНН, список счетов).
Статус платежа по ID.
Выписка по счёту за период (список транзакций). Поддерживает пагинацию.
Two tools (list_counterparties, get_company_info) registered without explicit inputSchema in registerTool call. Cannot verify parameter constraints or types.
No output schema documentation. LLMs cannot infer what fields to expect from tool responses, forcing them to parse unstructured text. Breaks downstream tool chaining (e.g., get_accounts returns account_id, but no schema confirms this).
Descriptions lack WHEN/WHY context. E.g., 'Баланс по счёту' (Balance by account) does not explain when to call this vs get_statement or summarize_transactions. Baseline: 194 chars; actual avg ~80 chars. LLMs struggle to select the right tool.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | 2026-07-28+ | v2 |
Список сохранённых контрагентов (получателей платежей).
Сводка по выписке: количество, суммы поступлений/списаний, чистый итог за период.
No error handling or recovery guidance. If create_payment fails (e.g., insufficient funds, invalid account), the response does not tell the LLM what to do next. No categorization of retryable vs fatal errors.
Parameter descriptions are terse (avg ~30 chars vs baseline 72). E.g., 'ID счёта' (Account ID) does not explain format, constraints, or how to obtain it. LLMs may pass invalid values.