Loading ToolBench
An MCP server that provides tools to interact with Google Workspace services including Gmail, Google Calendar, and Google Drive.
Server has 9 tools with complete schemas and descriptions. Naming follows verb_noun convention (read_, send_, list_, create_, update_, delete_). Descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameters are typed and described, but lack constraints (enums, ranges, patterns). Output schemas are partially documented via Pydantic models but not exposed in tool definitions. Error handling is minimal, most tools raise generic exceptions without recovery guidance. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Missing pagination for list tools. No idempotency or confirmation patterns for destructive operations.
Creates a new event in the user's primary Google Calendar.
Deletes a calendar event from the user's primary Google Calendar.
Retrieves the content of a file from Google Drive.
Lists calendar events within a specified time range, optionally filtering by a search query.
Lists files from the user's Google Drive, optionally filtered by a search query.
Searches for emails by subject and returns the body and snippet of the most recent matches.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels. delete_calendar_event is DESTRUCTIVE but unmarked; send_email is WRITE but unmarked. LLMs cannot distinguish safe reads from dangerous writes without annotations.
List tools (list_calendar_events, list_drive_files, read_email_by_subject) lack pagination parameters (limit, offset, page_size) and do not document result limits. Unbounded results risk context window exhaustion. Baseline: paginated-result pattern requires limit/offset and total count.
Destructive operations (delete_calendar_event) lack confirmation or dry-run support. Agents can delete events without verification, risking accidental data loss. No recovery guidance in error messages.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Reads the most recent email from the user's Gmail inbox.
Sends an email from the user's Gmail account.
Updates an existing calendar event with new details.
Parameter descriptions lack format constraints. 'start_time' and 'end_time' say 'ISO 8601 format' but do not specify timezone handling, whether seconds are required, or what happens if end_time < start_time. LLMs frequently miscalculate timestamps.
Error handling is minimal. Most tools raise generic exceptions ('API Gmail error', 'No emails found') without actionable recovery guidance. No distinction between retryable (network timeout) vs user-fixable (invalid subject) vs fatal errors.