Query Cursor Enterprise team usage, spending, and analytics. Requires a CURSOR_API_KEY from your team settings.
Server has 15 tools with consistent naming (verb_noun pattern), non-empty descriptions, and partial input schemas. Naming quality is strong across all tools (get_*, clear intent). Most tools have good descriptions (100-300 chars). However, critical gaps exist: (1) NO output schemas documented anywhere, tools return responses but callers don't know the field structure ahead of time. (2) Parameter validation and constraints are minimal, most date/string params lack format specifications or valid-value guidance. (3) Error handling is absent from visible code, no recovery guidance or error classification. (4) Some parameters could be more constrained (e.g., startDate/endDate accept flexible string formats but no enum/pattern enforcement visible in schema). (5) One tool (set_spend_limit) is a WRITE operation but has no confirmation/dry-run pattern or explicit warning. Overall, the server reads well at first glance but lacks production-grade rigor in output documentation, input validation, and error paths.
Get agent edits: suggests, accepts, rejects, and line statistics by color for diff suggestions.
Get billing groups with member lists, group-level spend, and daily spend breakdown. Also returns billing cycle dates.
Get client version distribution across the team: which versions are in use and their adoption percentages.
Get command usage analytics: frequency of cmd+k and other commands across the team.
Get daily usage data per user: lines added/deleted, applies, accepts, rejects, tabs, requests by mode (composer/chat/agent), models used, and client versions.
Get daily active users over time, including breakdowns for CLI, cloud agent, and Bugbot usage.
No output schemas documented. Tools return formatted text or JSON, but callers cannot infer field structure, types, or pagination shape ahead of time. This forces LLMs to parse unstructured responses, increasing hallucination risk and token waste.
Parameter format constraints are not enforced in schema. Date parameters accept flexible string formats ('YYYY-MM-DD', '7d', '30d', 'today', 'yesterday') but no regex pattern or enum is visible in the Zod schema definition. LLMs may pass invalid formats.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 56 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Get top file extensions by usage: suggestions, accepts, rejects, and line statistics per file type.
Get MCP tool usage breakdown: which tools are being called, from which servers, and how often.
Get model usage breakdown per day: which models are being used, how many messages, and by how many users. Essential for understanding model adoption and cost drivers.
Get plan usage breakdown: which models are being used under each plan (e.g., standard, pro, points).
Get current billing cycle spending for all team members. Shows spend in dollars, included vs overage, fast premium requests, and spend limits.
Get tab completion analytics: suggestions, accepts, rejects, and line statistics.
List all team members with their roles and status. Returns name, email, role, and whether they've been removed.
Get granular per-request usage events with model, token counts, costs, and whether the request was chargeable. Supports filtering by user email and date range.
Set a hard spending limit (in dollars) for a specific team member. Use with caution — this will block the user from making requests once the limit is reached.
set_spend_limit is a destructive WRITE operation (blocks user from making requests) but lacks a dry-run, confirmation, or explicit danger warning in the description. Agent could accidentally set limits without safeguards.
No error handling or recovery guidance visible in code. Tools throw on missing CURSOR_API_KEY, but no other error paths document how LLMs should handle API failures, rate limits, invalid pagination, or missing data.
Pagination support is inconsistent. Some tools support page/pageSize/allPages, others only handle startDate/endDate/page. No explicit documentation of pagination shape (total count, next_cursor, etc.) or limits on result size.
Parameters like 'email' and 'users' are strings but lack validation or format documentation. Should specify: valid email format, comma-separated parsing rules, case sensitivity.
Response formatting via formatResponse() wraps results in text blocks, which forces LLMs to parse JSON or tables from within a string. Structured tool output would be more efficient.