GitHub Copilot for JetBrains Admin Guide: Default Model Policy, Managed Settings, and MCP Startup Control

Introduction

If you administer GitHub Copilot for a team of JetBrains users, the October 10, 2026 plugin update is the most useful release of the year for you. It hands enterprise admins a managed setting to define the default Copilot model for every new conversation across the organization, adds a one-click Fix action for diagnostics, and introduces a switch to stop MCP servers from starting automatically for Copilot and Claude tools. This tutorial walks through the admin side: the managed settings file, the model default, the MCP startup control, and the developer experience that results.

Quick answer

The October 10, 2026 Copilot for JetBrains release lets enterprise admins lock a default agent model for all new conversations through managed settings while developers keep their per-session model picker override.

Evidence

The facts below come from GitHub’s own release notes and Copilot enterprise documentation, both published in October 2026.

The release. GitHub shipped the JetBrains plugin update on October 10, 2026. As the official changelog puts it: “This update brings more control over default models and MCP server in GitHub Copilot for JetBrains.” The same notes list the Fix action, the MCP startup switch, turn-grouped chat navigation, a Switch Account control, and reliability fixes across agent sessions, customizations, and worktrees. This is the primary source for every claim in this article, verified live on October 16, 2026.

The managed settings contract. The enterprise managed settings reference documents the top-level model key: setting it to a specific model ID makes that model the default for new conversations, and setting it to "auto" keeps automatic model selection as the default. Users can still select a different model on a per-conversation basis. The reference also notes that model was originally documented as the nested permissions.model key and that clients still honor the old nesting, but new configurations should use the top-level key.

Two dates that matter. First, the plugin now requires JetBrains IDE 2025.2 or later; support for JetBrains IDE 2025.1 ended with this release. Second, October 22, 2026 is the date GitHub’s default-enable policy begins flipping unconfigured generally available Copilot features to enabled for Business and Enterprise customers, a separate policy from the model default described here, but one every admin should review in the same session.

Why the default model matters financially. Automatic model selection spreads usage across models of very different per-token cost. Fixing the default to the cheapest model that fits the team’s work is a direct lever on premium-request spend. See the sandboxing tutorial below for the companion controls that cap what agents can actually run.

Prerequisites

• A GitHub Copilot Business or Enterprise subscription with admin access to the enterprise or organization settings.
• The managed settings mechanism: a JSON file the enterprise distributes to workstations. GitHub clients, including the JetBrains plugin, read it and apply admin keys automatically.
• JetBrains IDE 2025.2 or later on every workstation that should receive the new controls. Check Help -> About before troubleshooting anything else.
• Copilot for JetBrains plugin version from the October 10, 2026 release or later.
• A list of the model IDs your organization actually enables in the model policy, because a default model only works if the policy allows it.

Step-by-step guide

Step 1: Confirm the model policy allows your chosen default

Before you set any default, open the enterprise Copilot model policy and verify the model you want is enabled. The model policy decides availability; the managed setting decides the starting point. A default of claude-sonnet-5.5 does nothing if the policy disables that model for the org. Note the exact model ID from the policy page, because the managed setting takes a model ID string, not a display name.

Step 2: Add the managed model key

Edit the managed settings JSON that your enterprise delivers to IDEs and add the top-level model key:

{
  "model": "claude-sonnet-5.5"
}

To keep automatic model selection instead of a fixed model, use:

{
  "model": "auto"
}

This is the exact key contract from GitHub’s enterprise managed settings reference, which documents model as the replacement for the older nested permissions.model form. Distribute the file through your usual channel (GitHub’s docs describe the delivery mechanism for each enterprise deployment type), then have developers restart the IDE so the plugin re-reads the settings.

Step 3: Verify the default on a test workstation

On a machine that received the new settings:

1. Open any JetBrains IDE with the updated Copilot plugin.
2. Start a new Copilot chat conversation.
3. Look at the model picker: it should show the model you configured in the managed file.

If the picker shows a different model, check whether the developer made an explicit per-session choice before, since GitHub preserves explicit model picker choices over the admin default. Clear any saved conversation-level choice and open a fresh conversation to see the pure default.

For a per-team override, the managed settings reference supports specializing the model key on a per-team basis. A team that needs a different starting model can get its own settings file with "model": "unmanaged", which takes precedence over the enterprise file for members of that team.

Step 4: Disable automatic MCP server startup

The October 10 release adds a setting that disables automatic server startup for Copilot and Claude MCP tools. In the plugin’s MCP settings page, turn off automatic startup so configured servers initialize only when a developer explicitly starts or uses them.

This is a security and resource win: MCP servers can touch the filesystem and network on behalf of the agent, so an always-on server the developer did not ask for is attack surface and idle RAM. After flipping the switch, restart the IDE and confirm in the MCP settings that no server reports a running state until started manually.

Step 5: Teach the team the new Fix action

Diagnostics in the editor now carry a Fix intention-menu action. When a developer clicks it, the plugin opens inline chat and asks Copilot to repair the flagged issue, using agent mode when available and falling back to ask mode otherwise. The practical change: the old loop of reading a diagnostic, copying it into chat, and explaining the context collapses into one click. Teams measuring debugging throughput should expect routine fix-verification cycles to get shorter; tell developers that the action respects the current chat model, which is exactly the default you configured in Step 2.

Step 6: Roll out with the turn navigation and account controls

The same release groups chat messages by turn with fast navigation between turns, gives user messages copy and edit actions, and adds a clearer Switch Account control for people signed into more than one Copilot account. These are small, but they matter for support tickets: developers who share a machine or juggle a personal and a work account are the usual source of “my model is different” confusion, which is precisely what the Switch Account control and the clearer status display address.

Troubleshooting

Problem 1: A workstation does not get the new controls at all. Cause: the machine runs JetBrains IDE 2025.1 or older. The October 10 plugin update requires IDE 2025.2 or later and ended support for the 2025.1 branch. Fix: upgrade the IDE first (Help -> Check for Updates), then update the Copilot plugin from the Marketplace page. Confirm the IDE build number in Help -> About before retrying.

Problem 2: The managed default model does not apply to a developer. Cause: an explicit per-conversation model choice takes precedence over the admin default, by design. Fix: have the developer open a brand-new conversation without touching the picker and check the default there. Also verify the managed settings file actually reached that machine: check the plugin’s settings to see whether other managed keys are applying, and confirm the file distribution channel delivered the current version.

Problem 3: MCP servers still start automatically after disabling. Cause: the setting applies to Copilot and Claude tools in the plugin; a server configured at a different layer (for example, a project-level MCP configuration the plugin also reads) may start independently. Fix: list every configured server in the MCP settings page, note which ones report auto-start, and disable the ones you do not want. Restart the IDE after the change and re-check before declaring victory.

Problem 4: The Fix action opens ask mode instead of agent mode. Cause: agent mode is not available in that context. The changelog says the action uses agent mode when available and ask mode otherwise. Fix: confirm the developer’s plan and model support agent mode in the JetBrains plugin; if not, the ask-mode fallback is expected behavior, not a bug. The repair still happens, it just takes more turns.

Problem 5: “Model not available” errors after setting the default. Cause: the model policy does not enable the model you set as the default. The default and the policy are separate settings. Fix: go back to the enterprise model policy, enable the model (or pick a default that the policy already allows), and redistribute the settings.

Internal links

• GitHub Copilot Local Sandboxing Tutorial: the companion enterprise control for restricting what Copilot agents can touch on the filesystem, network, and macOS Keychain. Configure sandboxing before you widen agent mode usage with the Fix action.
• GitHub Copilot Computer Use Tutorial: if your team also gives Copilot desktop access, the sandbox and policy thinking carries over directly.
• MCP Server Tutorial: a primer on how MCP servers work, useful before you let teams add servers to the IDE plugin you just secured.
• GitHub Copilot Default Enablement (October 22): the admin checklist for the October 22 default-enablement change — confirm org-wide enablement before tightening per-IDE model defaults.

Conclusion

The October 10, 2026 JetBrains plugin update turns model choice from a per-workstation habit into an enforceable starting default, adds a one-click Fix path for diagnostics, and lets admins stop MCP servers from starting on their own. Configure the model key in managed settings, align it with the model policy, disable the MCP auto-start you do not need, and upgrade every machine to IDE 2025.2 or later. Next, harden what the agents can do with the GitHub Copilot local sandboxing tutorial, since a locked-down default model pairs best with locked-down tool permissions.

Frequently asked questions