A Model Context Protocol server for Google Calendar integration with auto authentication
Google Calendar MCP server has clear tool naming and decent parameter schemas, but descriptions are minimal and output structures are not documented. All 5 tools follow verb_noun naming convention (create_event, get_event, update_event, delete_event, list_events) which is good. Input schemas are present and use Zod with JSON Schema generation. However, tool descriptions are uniformly generic (10-50 chars), parameter descriptions lack context about format/constraints, and there is no documented output schema. The code shows solid implementation (OAuth handling, proper Google Calendar API integration) but the public interface lacks the LLM-optimized documentation that would let agents select and compose tools confidently. This server lands in the C/D range, functional but under-documented.
Creates a new event in Google Calendar
Deletes an event from the calendar
Retrieves details of a specific event
Lists events within a specified time range
Updates an existing event
Tool descriptions are too generic and lack LLM-optimization guidance. All descriptions are 10-50 characters, below the 100-char minimum recommended for agent tool selection. Example: 'Creates a new event in Google Calendar' tells the LLM WHAT but not WHEN to use it, prerequisites, or consequences.
No output schemas documented. LLMs cannot infer what fields (event_id, start_time, attendees, etc.) will be returned by create_event, get_event, or list_events. This forces agents to discover the response structure via trial-and-error, increasing failure rates and token waste.
Parameter descriptions lack actionable constraints. For example, 'summary' has description 'Event title' but no guidance on length limits, required characters, or forbidden values. 'dateTime' says 'ISO format' but does not clarify whether milliseconds are required, timezone handling, or what happens if the time is in the past.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 50 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 48 | - | v1 |
list_events lacks pagination documentation. The tool accepts 'maxResults' parameter but does not specify a default, min/max bounds, or whether a 'nextCursor'/'nextToken' is returned for continuation. Agents cannot safely paginate large event sets.
Destructive operations (delete_event) lack confirmation/dry-run capability and error recovery guidance. The tool description does not warn that deletion is irreversible or suggest a pattern to protect against accidental deletions by agents.
No error handling guidance in tool descriptions. LLMs will not know what to do if create_event returns an error (retry? check parameters? request user input?). Baseline: 'If authentication fails, check that gcp-oauth.keys.json is loaded. If a time conflict occurs, call list_events() to check availability.'
timeZone parameters are marked optional, but behavior is undefined if omitted. Will the API default to UTC? The user's local timezone? The calendar's default? This ambiguity will cause agents to pass incorrect or missing timezone information.
Parameter naming for 'orderBy' enum in list_events is inconsistent with Google Calendar API documentation. The enum accepts 'startTime' and 'updated' (camelCase) but does not explain what these mean or link to the underlying API semantics. Agents may guess wrong between 'startTime', 'start_time', and 'createdTime'.