Server has 8 tools with complete input schemas and descriptions. Naming follows verb_noun convention (create_, list_, read_, update_, delete_, add_, expire_). Descriptions are present (avg ~80 chars) but lack actionable context for LLM selection. Parameters are typed but many lack format constraints or validation guidance. Output schemas are not documented. Error handling is minimal, no recovery guidance or categorization. Security concerns: file paths accepted as parameters without sanitization; no permission gates on destructive operations.

• Document output schemas for all tools. Example: create_memento should return {memento_id: string, scope: string, created_at: ISO8601, attachments_count: number}. This enables LLMs to chain calls and extract IDs.
• Add enum constraints to scope parameter: {"type": "string", "enum": ["global", "repo"], "description": "Scope: 'global' (user-wide) or 'repo' (current repository)"}. Prevents hallucinated values.
• Validate file paths in copy_files parameter. Reject paths outside home directory or containing '..'. Return clear error: 'Path traversal not allowed. Paths must be absolute and within home directory.'
• Add dry-run support to delete_memento and expire_mementos. Return what would be deleted without actually deleting. Agents can then confirm before executing the real operation.
• Enhance descriptions with WHEN/WHY context. Example: 'Update an existing memento's title, body, tags, or expiration. Use this to modify a memento after creation. Returns the updated memento ID and new content hash for verification.'