Preflight Checklist
What's Wrong?
Setting disableAutoUpdates = true via MDM causes Claude Desktop to lock the entire third-party inference settings UI as "Organization-managed" — even though no inference-related keys are configured. Admins who only want update control lose all user ability to self-configure inference without deploying a full MDM inference profile.
Problem statement
Claude Desktop reads MDM preferences from the com.anthropic.claudefordesktop domain on macOS. The moment any key is present in that domain — even a single disableAutoUpdates = true — the app treats the device as fully managed and marks the complete third-party inference configuration UI as read-only with an "Organization-managed" banner.
This is an all-or-nothing behaviour. There is no way to tell Claude Desktop "only this key is managed; everything else remains user-configurable."
What Should Happen?
Expected behaviour
The UI lock should be key-scoped, not profile-scoped. If only disableAutoUpdates is deployed via MDM, only the auto-update setting should appear as "managed". All other settings — including the full third-party inference configuration — should remain fully editable by the user.
This is already how macOS system preferences work in general: individual plist keys can be managed while others remain free.
Suggested solution
Introduce per-key locking in the UI. Each settings section should check independently whether the relevant MDM keys are set — not whether any MDM profile exists at all. Example logic:
Workarounds (all unsatisfactory)
- Deploy a full inference configuration profile via MDM alongside the update key — high overhead for what should be a simple user-facing setting.
- Block update URLs at the network/firewall level instead of using MDM keys — brittle and bypasses Anthropic's intended MDM mechanism.
- Install Claude Desktop to
~/Applications instead of /Applications so the user has write access — removes fleet version control.
Error Messages/Logs
Steps to Reproduce
Steps to reproduce
- Deploy the following minimal
.mobileconfig profile to a macOS device via MDM (e.g. Jamf, Kandji, Mosyle, Intune):
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
<key>PayloadContent</key>
<array>
<dict>
<key>PayloadType</key>
<string>com.anthropic.claudefordesktop</string>
<key>PayloadIdentifier</key>
<string>com.anthropic.claudefordesktop.updates</string>
<key>PayloadUUID</key>
<string>A1B2C3D4-E5F6-7890-ABCD-EF1234567890</string>
<key>PayloadVersion</key>
<integer>1</integer>
<key>disableAutoUpdates</key>
<true/>
</dict>
</array>
<key>PayloadType</key>
<string>Configuration</string>
<key>PayloadIdentifier</key>
<string>com.yourorg.claude.updatepolicy</string>
<key>PayloadUUID</key>
<string>B2C3D4E5-F6A7-8901-BCDE-F12345678901</string>
<key>PayloadVersion</key>
<integer>1</integer>
</dict>
</plist>
- Open Claude Desktop on the managed device.
- Enable Developer Mode via Help → Troubleshooting → Enable Developer Mode.
- Navigate to Developer → Configure third-party inference.
- Observe that every section shows the banner:
"This configuration is managed by your organization. Contact your IT administrator to make changes."
- All fields are read-only. The user cannot configure any inference provider, gateway URL, or API key — despite no inference-related keys being present in the profile.
Claude Model
None
Is this a regression?
I don't know
Last Working Version
No response
Claude Code Version
Claude 1.5354.0 (9a9e3d) 2026-04-29T01:14:34.000Z
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
Impact: Any IT admin who deploys even a single MDM key (e.g. for update control) inadvertently breaks the self-service inference configuration flow for all users on managed devices. This is particularly painful in organisations that want update governance but do not (yet) use a centralised inference provider.
Preflight Checklist
What's Wrong?
Setting
disableAutoUpdates = truevia MDM causes Claude Desktop to lock the entire third-party inference settings UI as "Organization-managed" — even though no inference-related keys are configured. Admins who only want update control lose all user ability to self-configure inference without deploying a full MDM inference profile.Problem statement
Claude Desktop reads MDM preferences from the
com.anthropic.claudefordesktopdomain on macOS. The moment any key is present in that domain — even a singledisableAutoUpdates = true— the app treats the device as fully managed and marks the complete third-party inference configuration UI as read-only with an "Organization-managed" banner.This is an all-or-nothing behaviour. There is no way to tell Claude Desktop "only this key is managed; everything else remains user-configurable."
What Should Happen?
Expected behaviour
The UI lock should be key-scoped, not profile-scoped. If only
disableAutoUpdatesis deployed via MDM, only the auto-update setting should appear as "managed". All other settings — including the full third-party inference configuration — should remain fully editable by the user.This is already how macOS system preferences work in general: individual plist keys can be managed while others remain free.
Suggested solution
Introduce per-key locking in the UI. Each settings section should check independently whether the relevant MDM keys are set — not whether any MDM profile exists at all. Example logic:
Workarounds (all unsatisfactory)
~/Applicationsinstead of/Applicationsso the user has write access — removes fleet version control.Error Messages/Logs
Steps to Reproduce
Steps to reproduce
.mobileconfigprofile to a macOS device via MDM (e.g. Jamf, Kandji, Mosyle, Intune):Claude Model
None
Is this a regression?
I don't know
Last Working Version
No response
Claude Code Version
Claude 1.5354.0 (9a9e3d) 2026-04-29T01:14:34.000Z
Platform
Anthropic API
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information