MCP server connecting Claude to macOS Mail.app via AppleScript
Strong foundation with 20 well-named tools following verb_noun convention. All tools have descriptions (avg 156 chars, within 10-1024 baseline). Input schemas present and mostly complete with Zod validation. Key gaps: output schemas undocumented (no return type specs visible), error handling lacks recovery guidance, and some parameter descriptions could be more prescriptive about constraints. Composition is sound, tools are single-responsibility and chainable (e.g., list_messages → get_message → reply_to_message). Security: input sanitization via sanitize() function visible; no secrets in params. Naming is clear and consistent across domains (accounts, compose, mailboxes, messages).
Create a new mailbox (folder) in a Mail.app account. Can create top-level or nested mailboxes.
Delete a message from a mailbox.
Flag or unflag a message, optionally with a specific flag color.
Forward an existing email message to a new recipient. Returns the `sender` the forward was sent from. Does not support adding new attachments (Mail.app limitation).
Get detailed information about a specific Mail.app account by name
Get detailed information about a specific mailbox in a Mail.app account
Get the full details of a single message by its ID, including body, headers, recipients, and attachments. If mailboxName is omitted, searches all mailboxes in the account (slower but useful when you don't know which mailbox the message is in).
Output schemas not documented. No visible return type specifications for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., messageId from list_messages for chaining to get_message). Responses are JSON-stringified but structure is opaque.
Error handling lacks recovery guidance. toolError() function visible but no evidence of categorizing errors as retryable/user-fixable/fatal or providing actionable next steps. AppleScript failures will surface as generic errors without context for LLM retry logic.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
List all Mail.app accounts with their type, enabled status, and email addresses
List all attachments in a message.
List mailboxes in a Mail.app account, or all mailboxes across all accounts
List messages in a mailbox with pagination. Provide accountName and mailboxName to scope the listing.
Mark a message as read or unread.
Move a single message to a different mailbox.
Move multiple messages to a different mailbox in a single operation.
Read the contents of a text attachment inline. Supports .txt, .csv, .json, .html, .md, .xml, and .log files only.
Reply to an existing email message. Returns the `sender` the reply was sent from. Does not support attachments (Mail.app limitation).
Save all attachments from a message to a directory.
Save a single attachment from a message to disk.
Search for messages across mailboxes by field and query string. Supports filtering by date range, mailbox, and account.
Compose and send a new email message as plain text. Returns success when the message is queued for sending, along with the `sender` the message was sent from; actual delivery is not confirmed. Check Mail's Sent or Outbox mailbox to verify delivery.
Destructive operations (delete_message) lack confirmation/dry-run pattern. No evidence of a confirmation step before irreversible deletion. Agents can accidentally delete messages without safeguard.
Parameter descriptions for move_message, move_messages, flag_message, mark_read lack prescriptive constraints. E.g., flagIndex described as '-1 for default, 0-5 for specific colors' but no enum constraint or validation hint. LLMs may pass invalid values.
Pagination in list_messages and search_messages uses offset/limit but no total_count or next_cursor returned (inferred from schema only). Large result sets risk context window exhaustion without clear continuation signal.