14 tools with complete schemas and descriptions, but quality is uneven. Most tools are read-only with minimal parameter complexity. Naming follows verb_noun convention well (list_, get_, search_, set_, validate_). Descriptions are present but often terse (avg ~80 chars, below the 194-char baseline). Parameters lack depth: many accept optional account_id with minimal guidance on resolution behavior. Error handling is generic ('ok': False, 'error': str(exc)) without recovery guidance. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Output schemas are not formally documented, responses inferred from code. set_amp_account_state is the only write tool; it lacks confirmation/dry-run pattern. search_emails and list_mailboxes lack pagination (limit defaults to 20 but no cursor/offset or total_count). Attachment tools (get_email_attachment, read_email_attachment) have overlapping responsibility, could be clearer.

• Add tool annotations to all tools: @mcp.tool(readOnlyHint=True) for read-only tools, @mcp.tool(destructiveHint=True) for set_amp_account_state. This enables LLMs to reason about safety.
• Enhance error responses with recovery guidance. Instead of {'ok': False, 'error': 'Account not found'}, return {'ok': False, 'error': 'Account not found', 'recovery': 'Call list_email_accounts() to see available accounts', 'retryable': False}.
• Add pagination to list tools: include 'total_count', 'next_cursor', and accept 'cursor' or 'offset' + 'limit' parameters. Document in descriptions: 'Returns up to 20 results; use cursor for pagination.'
• Expand descriptions to 100-200 chars. Add WHEN to use and prerequisites. E.g., 'list_known_contacts: Returns historical recipients from Sent folder for the account. Use this to populate recipient suggestions or validate email addresses before sending.'
• Add dry-run support to set_amp_account_state: accept 'dry_run=True' parameter. Return proposed changes without committing. This prevents accidental state mutations.