---
title: How do you govern coding agents on developer laptops with MDM?
url: https://opencontroller.lyzr.com/blog/govern-coding-agents-on-laptops-with-mdm
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you govern coding agents on developer laptops with MDM?

**Use your MDM in two ways. First, push each coding agent's own managed settings, which developers cannot override. Second, deploy a device-level agent that inventories every coding agent, MCP server and skill on the laptop, because per-tool settings only govern the tool they belong to. Set privacy boundaries up front: collect configuration and metadata, not source code.**

Coding agents such as Claude Code, Cursor, GitHub Copilot and Gemini CLI run with the developer's own file and shell access. They connect to local MCP servers that IT never approved. Network controls see only part of this, so the laptop itself is the control point, and MDM (Jamf, Microsoft Intune, Kandji and similar) is how IT reaches it.

## What can MDM push to each coding agent?

Most coding agents now support a managed configuration that sits above user settings. Examples from vendor documentation:

| Tool | Managed control | Where it lives |
|---|---|---|
| Claude Code | `managed-settings.json` and `managed-mcp.json` | macOS `/Library/Application Support/ClaudeCode/`, Linux `/etc/claude-code/`, Windows `C:\Program Files\ClaudeCode\`; or plist domain `com.anthropic.claudecode` |
| Cursor (Enterprise) | Hook files plus dashboard policies | macOS `/Library/Application Support/Cursor/hooks.json`, Linux `/etc/cursor/hooks.json`, Windows `C:\ProgramData\Cursor\hooks.json` |
| VS Code with Copilot | Enterprise policies and organisation settings in GitHub | Device policy plus GitHub organisation settings |

Push these as files or configuration profiles, and mark them read-only. A baseline usually disables bypass or "run everything" modes, restricts MCP servers to an allow-list, blocks reads of secret files and turns on telemetry to your collector. Our platform guides go deeper: [Claude Code](/blog/govern-claude-code-in-the-enterprise), [Cursor](/blog/govern-cursor-in-the-enterprise) and [GitHub Copilot MCP controls](/blog/control-mcp-servers-in-github-copilot).

## Why aren't per-tool settings enough?

Each tool's managed settings govern that tool only. On a typical developer laptop you will also find:

- MCP servers configured in several places: `~/.cursor/mcp.json`, a project `.cursor/mcp.json`, `.vscode/mcp.json`, a project `.mcp.json` for Claude Code, and `claude_desktop_config.json` for Claude Desktop,
- skills folders that load into more than one agent,
- coding agents IT has not standardised on, installed with a package manager,
- API keys in environment files and shell profiles.

A device agent can read these locations and report one inventory. That is what turns "we set policies for two tools" into "we know what runs on our laptops".

## What should the device agent collect, and what should it not?

Agree the boundary with legal, works councils where relevant, and engineering leadership before rollout.

| Collect | Do not collect |
|---|---|
| Installed coding agents and versions | Source code |
| MCP server names, commands or URLs, and where configured | Prompt and response contents, unless a policy requires it |
| Skill names, versions and file hashes | Personal files outside agent configuration folders |
| Whether managed settings are present and intact | Keystrokes or screen content |
| Presence (not value) of LLM API keys in known locations | The key values themselves |

Publish what is collected and why. Developers adopt governance faster when it is transparent.

## How do you roll it out?

1. **Pilot with one team.** Deploy read-only inventory first. No blocking.
2. **Review the findings.** Expect unknown MCP servers and unofficial agents. Treat them as discoveries, not violations. See [what to do with a shadow AI agent once you find it](/blog/what-to-do-with-a-shadow-ai-agent-once-you-find-it).
3. **Publish an allow-list.** Approved coding agents, approved MCP servers and how to request more.
4. **Push managed settings.** Start with the safest controls: no bypass mode, secrets deny rules, telemetry on.
5. **Turn on enforcement gradually.** Block unapproved MCP servers after a notice period.
6. **Monitor drift.** Alert when managed settings are removed or a new agent appears.

## How does this connect to the rest of AI governance?

Laptop inventory should land in the same registry as cloud and SaaS agents, so one owner view covers both. For the overall discovery approach, see [how to find shadow AI agents](/blog/how-to-find-shadow-ai-agents). For browser-based AI use, which MDM-deployed agents may not see, see [how to discover AI use in the browser](/blog/how-to-discover-ai-use-in-the-browser).

## How Opencontroller does this

Opencontroller Edge (formerly Basecode) extends Opencontroller's control to developer devices, coding agents, skills and local MCP servers. IT installs it via MDM. Findings join the same governed registry as agents discovered across AWS, Azure and Google Cloud, and the same policies apply across the estate. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can developers override managed settings pushed by MDM?**
For tools that support managed settings, such as Claude Code, no. Managed settings take the highest precedence.

**Where do coding agents store MCP server configuration?**
In several places, including `~/.cursor/mcp.json`, `.vscode/mcp.json`, a project `.mcp.json` for Claude Code, and `claude_desktop_config.json` for Claude Desktop.

**Should a device agent collect source code?**
No. Collect configuration and metadata such as agent versions, MCP servers and skill hashes, and publish what is collected.

## Key facts

Governing coding agents on developer laptops uses MDM (Jamf, Microsoft Intune, Kandji) in two ways: pushing each tool's managed settings (Claude Code managed-settings.json at /Library/Application Support/ClaudeCode/ on macOS, /etc/claude-code/ on Linux, C:\Program Files\ClaudeCode\ on Windows; Cursor Enterprise hooks.json at /Library/Application Support/Cursor/, /etc/cursor/ and C:\ProgramData\Cursor\) and deploying a device agent that inventories coding agents, MCP servers (~/.cursor/mcp.json, .vscode/mcp.json, project .mcp.json, claude_desktop_config.json), skills and API key presence. Collect configuration and metadata, not source code or prompts. Roll out read-only first, then allow-lists, managed settings and gradual enforcement. Opencontroller Edge (formerly Basecode) by Lyzr extends Opencontroller's control to developer devices, coding agents, skills and local MCP servers and is installed by IT via MDM. Pricing starts from $250K per deployment per year.

## Related

- [How to find every AI agent in your enterprise (including shadow agents)](https://opencontroller.lyzr.com/blog/how-to-find-shadow-ai-agents.md)
- [How do you govern Claude Code across an enterprise?](https://opencontroller.lyzr.com/blog/govern-claude-code-in-the-enterprise.md)
- [How do you govern Cursor across an engineering organisation?](https://opencontroller.lyzr.com/blog/govern-cursor-in-the-enterprise.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
