Attestari provides 8 tools with complete input schemas and descriptions. Naming follows verb_noun convention (add_memory, search_memory, get_timeline, etc.). Descriptions are substantive (100-200 chars), explaining WHAT and WHEN. However, output schemas are not documented in the MCP tool definitions, the server code shows FastAPI endpoints returning structured responses, but the MCP layer does not expose return type documentation. Parameters are well-described with types and constraints. Error handling is implicit (HTTP status codes) but not surfaced as recovery guidance in tool descriptions. Security-sensitive operations (forget, resolve_entities) lack explicit confirmation/dry-run patterns in the schema, though the code supports dry_run for forget.

• Document output schemas for all 8 tools in the MCP tool definitions. Include field names, types, and examples. Reference the FastAPI response models (e.g., DeletionCertificate for forget, Edge for search_memory).
• Enhance forget and resolve_entities descriptions to emphasize dry_run and confirmation patterns: 'Use dry_run=true to preview changes before committing. This is a destructive operation, verify the subject_id before executing.'
• Add error recovery guidance to all tool descriptions. E.g., 'If subject_id is not found, try search_memory() with a partial query to locate the correct subject.'
• Clarify parameter dependencies in descriptions. E.g., 'subject_id and subject are mutually exclusive; provide one or the other. If you only have a name, use subject; if you have a system ID, use subject_id.'
• Document pagination and result limits in search_memory and get_timeline. E.g., 'Returns up to limit results (default 5). Use as_of to time-travel to past states. For large result sets, paginate via limit and offset.'