Copilot for JetBrains: MCP Startup and Model Controls

GitHub updated Copilot for JetBrains on October 10, 2026. Enterprise administrators can now choose the default agent model for new conversations, while developers retain the model picker. A separate setting can prevent configured Copilot and Claude MCP servers from starting automatically. Diagnostic menus also gain a Fix action that opens inline chat, using agent mode when available and ask mode otherwise. The release ends support for JetBrains IDE 2025.1: version 2025.2 or later is required.

Why these controls are different

A default model is a starting preference, not a model lock. MCP startup determines when configured tool servers activate, not which external systems their credentials can reach. GitHub documentation separately requires organizations on Copilot Business or Enterprise to enable the MCP servers in Copilot policy. Teams should therefore review model defaults, automatic startup, organization policy and individual tool permissions as separate decisions. None of these changes alone proves that an agent session is isolated or that an integration is safe.

Five checks before a team-wide rollout

1. Inventory the JetBrains IDE versions on developer machines. Upgrade installations still on 2025.1 before adopting the updated plugin, and confirm that each environment runs 2025.2 or newer.
2. Pilot the current Copilot plugin with a small group. Test sign-in, account switching, model selection, chat navigation and existing agent sessions in the actual IDEs your team uses.
3. Choose a sensible enterprise-managed default agent model for new chats. Tell developers that the model picker remains available, so the setting is a starting point rather than an enforced single-model policy.
4. Review configured MCP servers and the organization policy that governs them. Decide whether automatic startup is appropriate for each environment, then test explicit startup and tool access using a low-risk repository.
5. Try the new diagnostic Fix action on a reproducible noncritical problem. Inspect the suggested change and run normal tests before applying it broadly; a proposed AI fix still needs developer review.

What the announcement does not establish

GitHub reports reliability improvements across language-server startup, model and provider switching, MCP configuration, customization refreshes, file changes and worktrees. It does not provide a measured productivity gain for this particular update. Nor does the new MCP startup preference replace credential scoping, repository access rules or human approval. Treat those as separate controls and measure tool activation, session errors and developer friction during the pilot instead of assuming every workflow is automatically safer or faster.