Cursor audit logs: what admins can see, by plan, and the gaps

the short answer

Partly. As of October 2026, Cursor audit logs are an Enterprise feature that records admin events such as sign-ins, roles, settings, hooks and MCP configuration, and not prompts, agent output or code. Enterprise OpenTelemetry export adds token, cost and tool-call metrics, but conversation content covers Cloud Agents and Grok Bot only. For a record of what IDE and CLI agents actually ran, Cursor points to hooks.

Audit log plan: Enterprise only
Audit log content: Admin and configuration events; no prompts, agent output, code or credentials, except team_hook events, which include the hook’s script or prompt text
API window: Last 7 days by default, at most 30 days per query, 500 events per page
Total retention: Not published by Cursor
Prompts and actions: Via hooks; OTel conversation content covers Cloud Agents and Grok Bot only

Cursor Log Sources Compared

Cursor’s own compliance page says audit payloads carry identifiers, names and changed settings, “never prompts, agent output, generated code, or credentials”; the one payload with free text is team_hook, which includes the hook’s script or prompt. Non-human actors appear as Api Key: <name>, Bot: <owner> or SCIM. The API returns at most a 30-day window per query; Cursor does not publish total retention, so export to your SIEM if you need more.

Source	Plan	What it records	Gaps
Audit log (dashboard, CSV, `GET /teams/audit-logs`, `GET /organizations/audit-logs`)	Enterprise	Sign-ins, membership, roles, API keys, settings, rules, hooks, MCP config, privacy mode changes; `user_email` per event	No prompts, agent output or code; failed sign-ins and Cloud Agent automation runs not logged
SIEM streaming (Splunk HEC, Sumo Logic, HTTPS webhooks, S3)	Enterprise, set up by emailing Cursor	The same audit events	Same as the audit log
OpenTelemetry export	Enterprise	`cursor.token.usage`, `cursor.tool.calls`, `cursor.cost.usage`, model request and hook logs	Conversation content (Cloud Agents and Grok Bot only) is opt-in and blocked under Privacy Mode (Legacy); the other telemetry families default on for a new destination
Usage APIs (`daily-usage-data`, `filtered-usage-events`, `spend`)	Teams	Usage and spend per member	Not an action record
Hooks (`beforeShellExecution`, `beforeSubmitPrompt` and others)	Any	Whatever your hook script logs: commands, prompts, file reads	You build and run the pipeline

Which Source Answers Which Question

Question	Best source
Who changed hooks, rules or MCP config?	Audit log
Who turned privacy mode off?	Audit log (`privacy_mode` events)
How many tool calls and tokens per developer?	OTel export or usage APIs
What did the agent run on this laptop yesterday?	Hook logs, or a hook-level policy client
What did the developer ask the agent?	A `beforeSubmitPrompt` hook (IDE and CLI); OTel content only for Cloud Agents

Filling the Gap with a Hook-Level Client

Failproof hooks Cursor (blocking verified on PreToolUse, UserPromptSubmit and Stop) and records each agent session and policy decision per machine, in the same format as Claude Code, Codex and Copilot CLI sessions. Before connecting anything, failproofai audit scans the Cursor histories already on a laptop offline and shows risky patterns. failproofai config installs hooks at user scope (Failproof supports user and project scope for Cursor). Copying the same hook entry into Cursor’s enterprise hooks.json with failClosed: true is a manual step outside what Failproof installs; test it on one machine before rolling it out. See the Cursor hooks example.

FAQ

Can Cursor admins see what employees prompt?

Not in the audit log. Conversation content in the OTel export covers Cloud Agents and Grok Bot only as of October 2026; for IDE and CLI prompts, use a beforeSubmitPrompt hook. See can admins see prompts.

Is there an audit log on the Teams plan?

No. Audit logs, SIEM streaming and OTel export are Enterprise features.

How long does Cursor keep audit logs?

Cursor does not publish total retention. The API serves at most 30 days per query, so stream or export to your own store.

Sources

Checked against the sources below on October 21, 2026. Harness settings, hook capabilities and cloud features change between versions.