elim-mcp demonstrates solid tool design with clear naming (verb_noun pattern), comprehensive parameter schemas using Zod, and well-structured descriptions. All 4 tools have explicit registrations with inputSchema definitions. Descriptions are domain-specific and actionable (avg ~150 chars). However, output schemas are not formally documented, responses are formatted as text rather than structured objects, limiting downstream tool chaining. Error handling returns text messages but lacks categorization (retryable vs fatal). No tool annotations (readOnlyHint/destructiveHint) despite clear read/write semantics.

• Add formal output schemas to each tool. Define structured return types (e.g., { records: [{id, kind, topic, content, created_at, scope}], total_count, next_cursor }) instead of formatted text strings. This enables reliable field extraction and downstream chaining.
• Implement tool annotations: add readOnlyHint: true to recall/get_current/trace; add destructiveHint: true to remember. This signals to agents which tools have side effects.
• Categorize errors with structured responses: { error: string, category: 'retryable'|'user_fixable'|'fatal', suggestion: string }. Example: 'Invalid UUID format, must be a valid record_id. Try recall() to find existing records.'
• Add pagination to recall and get_current: accept optional limit (default 20, max 100) and offset/cursor parameters. Return total_count and next_cursor in response so agents can iterate large result sets.
• Rename 'get_current' to 'list_ledger_records' or 'get_ledger_thread' to clarify that it returns multiple records for a scope, not a single 'current' record.