Server has 11 tools with basic naming and descriptions, but significant gaps in schema completeness, parameter documentation, and error handling. Most tools lack detailed input/output schemas. Descriptions are present but often generic (10-50 chars). No parameter descriptions for most inputs. No error recovery guidance. Stateful design (state.json) violates modern MCP patterns. Cabinet tools rely on lazy imports and undocumented credentials handling.

• Document output schemas for all tools. For miit_get_timetable, specify: {timetable: [{date, time, subject, room, instructor}], semester: string, date_range: string}. For cabinet_get_profile, specify: {full_name, email, phone, address, ...}.
• Add parameter descriptions explaining format and constraints. E.g., group_id: 'Numeric timetable ID (e.g., 12345) obtained from miit_list_institutes or miit_find_group. If omitted, uses stored group from miit_set_current_group.'
• Implement stateless request handling. Move group context from state.json to request parameters or session metadata. Accept group_id in every tool that needs it; remove implicit state dependency.
• Add error handling with recovery guidance. Wrap network calls in try-catch; return structured errors like {error: 'group_not_found', message: 'Group УВП-171 not found. Call miit_list_institutes to see available groups.', retryable: false}.
• Document cabinet credentials handling explicitly. Specify: 'Requires MIIT_LOGIN and MIIT_PASSWORD environment variables or credentials.json in server directory. Never pass credentials as tool parameters.' Add validation to reject credential parameters if passed.