GitHub Copilot for JetBrains IDEs now lets enterprise administrators set the default agent model for new conversations through managed settings, and adds a setting that stops MCP servers starting automatically. For teams, the default model is a starting point rather than a lock, and the MCP setting decides when connected tools can act.
What was released
The update was announced by GitHub on 10 October 2026 in the GitHub changelog, under the Copilot and enterprise management tools labels. It ships in the GitHub Copilot plugin, available from the JetBrains Marketplace. The same entry ends support for JetBrains IDE 2025.1, so the plugin now needs JetBrains IDE 2025.2 or later.
What actually changed
According to the announcement, three changes matter for teams running agents in the IDE:
- Enterprise-managed default models. Enterprise administrators can use managed settings to choose any available Copilot agent model as the default for new conversations. Developers can still pick a different model in the model picker.
- MCP server startup control. A new MCP setting turns off automatic server startup for Copilot and Claude, so configured tools become active only when someone starts them.
- Fix from diagnostics. The diagnostic intention menu now has a Fix action that opens inline chat and asks Copilot to repair the issue, using agent mode when it is available and ask mode otherwise.
The rest of the update is usability work: clearer account labels and a Switch Account action, conversation messages grouped by turn, copy and edit on your own messages, and reliability fixes for language server startup, model switching, MCP configuration and worktree workflows. GitHub does not state plugin version numbers, preview status or plan availability for these features in the entry.
A default is not a policy
The managed default sets which model a new conversation opens with. It does not stop a developer choosing another model, so it is a way to steer usage, not a control. That is still useful: most people keep the default, so it shapes cost and behaviour across the organisation without a memo.
If you need a hard limit on which models can be used, the IDE default will not give you one, so look to the admin controls on your Copilot plan and check what they cover. Treat the two as separate decisions: which models are allowed, and which allowed model new work starts on. Our guide on controlling AI agent running costs covers how model choice drives spend.
Why MCP startup matters
An MCP server that starts with the IDE gives the agent its tools before anyone has looked at what is configured. Some of those tools reach internal systems, run commands or read data. The new setting means a server can sit in the configuration but stay inactive until a developer chooses to start it.
In CodeDTX's view, that is the safer default for repositories where agents could reach production data or credentials. It pairs with the questions in how AI agents connect securely and stopping agents reaching restricted data: decide which servers are allowed at all, then decide whether they start on their own.
What to set before rolling it out
- Upgrade IDEs first. Anyone still on JetBrains IDE 2025.1 loses plugin support, so check the IDE versions in use before announcing the new controls.
- Pick the default model deliberately. Choose it for the work most developers do, balancing cost and quality, and write down why so the choice can be revisited when GitHub deprecates models.
- Turn off automatic MCP startup where tools touch sensitive systems. Keep auto-start only for servers that are read-only or low risk.
- Review what the Fix action can change. It uses agent mode where available, so a quick fix can edit files. Keep normal code review in place; Copilot code review and human review still apply.
- Keep tool access governed in one place. The IDE settings sit alongside, not instead of, local sandboxing and your own agent inventory.
When not to change anything
If your organisation is not on a Copilot enterprise setup with managed settings, the default model control does not apply to you, and the plugin update is mostly usability. If developers do not use MCP servers in JetBrains, the startup setting changes nothing. And if a team relies on a specific model for reproducible results, do not move its default without testing the new model on the same tasks first.
Frequently asked questions
Can admins force a single model in Copilot for JetBrains?
Not with this setting. GitHub says enterprise administrators can use managed settings to choose any available Copilot agent model as the default for new conversations, but developers can still choose a different model in the model picker. The default steers which model new work starts on. Limiting which models can be used at all is a separate decision, so check the admin controls on your Copilot plan rather than relying on the IDE default.
Which JetBrains IDE version does Copilot now require?
GitHub has ended support for JetBrains IDE 2025.1 in the Copilot plugin. You now need JetBrains IDE 2025.2 or later to use the plugin, according to the 10 October 2026 changelog entry. Check the IDE versions across your team before relying on the new controls, because developers on the older release will not get them and lose plugin support.
What does the new MCP startup setting do?
It lets you turn off automatic startup of MCP servers for Copilot and Claude in JetBrains IDEs. Servers stay in the configuration but only become active when someone starts them. That gives teams a moment to check which tools an agent will be able to use, which matters most where servers reach internal systems, run commands or read sensitive data.
Does the Fix action change code on its own?
The Fix action in the diagnostic intention menu opens inline chat and asks Copilot to repair the issue. GitHub says it uses agent mode when available and ask mode otherwise, so in agent mode it can propose and apply edits. Treat its changes like any other agent output: review the diff, run the tests and keep normal code review before merging.
