---
title: How do you govern Cursor across an engineering organisation?
url: https://opencontroller.lyzr.com/blog/govern-cursor-in-the-enterprise
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you govern Cursor across an engineering organisation?

**Govern Cursor through its admin dashboard and device policies. Enforce SSO and Privacy Mode, restrict models and MCP servers with allowlists, lock down agent run modes, push team rules and hooks, and pull audit and usage data through the Admin API. Most of the strongest controls need the Enterprise plan, and local MCP servers on laptops still need device-level discovery.**

Cursor is an AI code editor whose agent can run shell commands, call MCP tools and edit files. Governance has to cover what the agent can run, what it can reach and what data leaves the device.

## Which Cursor controls are on which plan?

| Control | Teams | Enterprise |
|---|---|---|
| SAML SSO (Okta, Azure AD, Google Workspace, OneLogin) | Yes | Yes |
| Enforce Privacy Mode org-wide | Yes | Yes (on by default) |
| Team Rules from the dashboard | Yes | Yes |
| SCIM 2.0 provisioning | No | Yes |
| Audit logs and service accounts | No | Yes |
| Model access restrictions and repository blocklist | No | Yes |
| MCP allowlist, run-mode and network controls | No | Yes |
| Team hooks pushed from the cloud | No | Yes |
| Analytics API and OpenTelemetry export | No | Yes |

Source: Cursor's enterprise documentation. Check your contract, because plan details change.

## How do you control what the Cursor agent can run?

Cursor's run modes decide whether the agent asks before running commands. The current modes are Auto-review, Allowlist and Run Everything. Cursor deprecated "Ask Every Time" in version 3.5 in May 2026 and made Auto-review the recommended default in version 3.6.

On Enterprise, admins can restrict which modes users see, and team settings override a developer's local `permissions.json`. The sandbox uses Seatbelt on macOS and Landlock or Bubblewrap on Linux, with network blocked by default. Extra protections cover the browser, file deletion and files outside the workspace.

A practical baseline:

1. Hide Run Everything from all users.
2. Use Allowlist for teams that work on production code.
3. Keep sandbox networking blocked, and allow domains per team only when needed.
4. Turn on file-deletion and external-file protection.

## How do you control MCP servers in Cursor?

Developers add MCP servers in `.cursor/mcp.json` in a project or `~/.cursor/mcp.json` globally. Without controls, any developer can connect any server.

On Enterprise, the MCP allowlist matches servers by command or URL, with `*` wildcards. Precedence is the dashboard, then `~/.cursor/permissions.json`, then editor settings. Higher sources replace lower ones rather than merging. Allowlisting a server does not install it. Each server can also have its own network setting.

The gap: an allowlist blocks unknown servers inside Cursor, but it does not tell you which servers developers already run, or what other tools on the same laptop use. Our guide to [reviewing MCP servers agents already use](/blog/how-to-review-mcp-servers-agents-already-use) covers that audit.

## What do hooks and team rules add?

Hooks run scripts on agent events such as `beforeShellExecution`, `beforeMCPExecution`, `beforeReadFile`, `beforeSubmitPrompt` and `afterFileEdit`. Exit code 2 blocks the action.

- Enterprise hook files live at `/Library/Application Support/Cursor/hooks.json` on macOS, `/etc/cursor/hooks.json` on Linux and `C:\ProgramData\Cursor\hooks.json` on Windows.
- Team hooks are pushed from the cloud and sync about every 30 minutes.
- Priority is Enterprise, then Team, then Project, then User. Deny beats ask, and ask beats allow.
- Hooks fail open by default. Set `failClosed: true` for any hook that enforces security.

Team Rules are instructions set in the dashboard and can be made required. They guide the agent but are not enforcement. Use hooks and run modes for anything that must not happen.

## How do you deploy and audit Cursor?

Deploy with MDM. On macOS, Cursor uses a configuration profile. On Windows, it ships ADMX and ADML Group Policy templates. Policies include `AllowedTeamId`, which stops developers signing in to a personal team, and `AllowedExtensions`. On Linux the policy file sits in the user's home folder at `~/.cursor/policy.json`, so it is weaker.

For audit, Enterprise audit logs are in the dashboard with CSV export and through the API, with a 7-day default window and 30 days maximum per call. Payloads exclude prompts and code. The Admin API covers members, daily usage, spend and per-user spend limits. The AI Code Tracking API, still alpha, splits lines per commit into Tab, Composer and non-AI. Export logs to your SIEM, because the API window is short.

For spend by team, see [set budgets and alerts for AI agent spend](/blog/set-budgets-and-alerts-for-ai-agent-spend).

## What does Cursor governance not cover?

Cursor governs Cursor. The same developer may also run Claude Code, GitHub Copilot and local MCP servers. Compare our [Claude Code governance guide](/blog/govern-claude-code-in-the-enterprise) and the [cross-platform guide](/blog/how-to-govern-ai-agents-across-platforms). An [AI agent control plane](/blog/what-is-an-ai-agent-control-plane) gives one view across them.

## How Opencontroller does this

Opencontroller Edge extends Opencontroller's control to developer devices, coding agents, skills and local MCP servers. IT installs it via MDM, so coding agents such as Cursor join the same governed registry as cloud agents. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can admins stop developers adding MCP servers in Cursor?**
On Enterprise, yes. The MCP allowlist matches servers by command or URL, and dashboard settings override local files.

**Do Cursor audit logs include prompts or code?**
No. Audit log payloads exclude prompts and code. They record admin and account events.

**Can Cursor hooks block risky commands?**
Yes. A hook on `beforeShellExecution` that exits with code 2 blocks the command. Set `failClosed: true` so a broken hook blocks rather than allows.

## Key facts

Cursor Teams supports SAML SSO, enforced Privacy Mode and Team Rules; Enterprise adds SCIM, audit logs, model restrictions, MCP allowlists, run-mode controls, cloud-pushed hooks and Analytics API. Run modes are Auto-review, Allowlist and Run Everything (Ask Every Time deprecated in v3.5, May 2026). Hooks block with exit code 2 and fail open unless failClosed is true. Audit log API window is 7 days default, 30 max; payloads exclude prompts and code. MDM policy AllowedTeamId stops personal-team sign-in. Opencontroller Edge (by Lyzr) covers coding agents and local MCP servers on devices via MDM. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern Claude Code across an enterprise?](https://opencontroller.lyzr.com/blog/govern-claude-code-in-the-enterprise.md)
- [How do you review MCP servers your agents already use?](https://opencontroller.lyzr.com/blog/how-to-review-mcp-servers-agents-already-use.md)
- [How do you get visibility of GitHub Copilot coding agent activity across your organisation?](https://opencontroller.lyzr.com/blog/visibility-of-github-copilot-coding-agent-activity.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
