JetBrains Copilot Can Now Wait to Start Your MCP Servers

At a glance

• Copilot for JetBrains now lets an enterprise set the default agent model, and lets each developer stop Copilot and Claude MCP servers from starting with the IDE.
• The same plugin update drops an older JetBrains IDE line, so pinned IDE builds need a check before the plugin updates.
• Copilot local sandboxing reached general availability after our pre-release coverage, which makes a required sandbox policy a realistic default.
• Every command-line channel we track held still, and only a Gemini CLI nightly appeared.

The weekend was quiet on the command line, which leaves the editor as the place a surprise is most likely this morning. GitHub's JetBrains update hands two separate decisions to two separate people: administrators choose which model a new conversation starts on, and developers choose whether configured MCP servers wake up on their own. Add the generally available sandbox from last week and you have the outline of a Monday policy review rather than a Monday upgrade. Treat today as the day to check which JetBrains IDE builds your team runs, decide whether MCP servers should start on their own, and confirm local sandboxing is on wherever Copilot runs commands for you.

Top Stories

1. Copilot for JetBrains adds a managed default model and an MCP startup switch

GitHub published "New controls and chat improvements in Copilot for JetBrains" at 2:53 PM ET on Saturday 10 Oct. Enterprise administrators can now use managed settings to choose any available Copilot agent model as the default for new conversations, and GitHub says a developer's explicit choice in the model picker is still preserved. A new MCP setting turns off automatic server startup for both Copilot and Claude, so configured servers become active only when you start them. Diagnostic intention menus gain a Fix action that opens inline chat and asks Copilot to repair the problem, using agent mode where it is available and ask mode otherwise.

The rest of the post is smaller but touches daily use. Signed-in accounts get clearer labels and a Switch Account action, chat messages are grouped by turn with controls to move between turns, and your own messages can be copied or edited. GitHub also lists reliability fixes for language server startup, model and provider switching, MCP configuration, customization refreshes and worktree workflows. The update ends support for JetBrains IDE v2025.1, so the plugin now needs v2025.2 or later. The post names no setting keys and gives no plugin version number, so we are not printing either here; look for the options in the plugin's own settings pages.

Practical dev impact: If you administer Copilot for an enterprise, the managed default is the cleanest way to start every JetBrains conversation on a model you have already priced and approved, while developers can still switch for a single chat. If your MCP servers hold credentials or reach production data, our read is that turning off automatic startup is worth doing today, because an IDE launch will no longer bring those servers up quietly in the background. Anyone still on JetBrains IDE v2025.1 has to upgrade the IDE before the plugin will keep updating.

2. Local sandboxing for Copilot is generally available, after our Wednesday brief

GitHub announced at 11:46 AM ET on Wednesday 7 Oct that local sandboxing is generally available in Copilot CLI, the Copilot app and VS Code sessions that use Agent Host. Commands that Copilot starts run with restricted access to the filesystem, network, credentials and other system capabilities, according to policies set by the developer or the organization. GitHub says the feature runs on Microsoft eXecution Container, which turns a single sandbox policy into native controls on Windows, macOS and Linux. Policies can limit which directories agent-run commands may read or change, control access to the internet, local networks, Git credentials and GitHub CLI credentials, and extend to local MCP and language servers where supported. Enterprise-managed settings can require sandboxing and stop developers from weakening it, at no extra cost.

This needs reconciling with our own coverage. Our Wednesday brief reported Copilot CLI v1.0.93-4, a pre-release GitHub published at 1:34 AM ET that day, whose notes opened command sandboxing to all users through /sandbox and --sandbox while the docs still called it experimental. That brief was written before the 11:46 AM ET GA post, so its advice to keep fleets off the pre-release reflected the state of things at the time and no longer describes the feature's status. Thursday's brief covered VS Code v1.141 treating the old sandbox device policy as an overridable default in the Copilot harness, and none of our posts since then has covered the GA announcement itself.

Practical dev impact: General availability is the point where requiring sandboxing through enterprise-managed settings becomes a sensible default instead of an experiment, and for VS Code teams it is also the answer to the override gap from Thursday. The Copilot CLI v1.0.95 sandbox credential injectHosts keys, covered in "Read the Large File Once. Compact the Subagent Early. Stable Stepped to v2.1.287.", sit on top of this sandbox, so our suggestion is to turn the sandbox on first and add host rules afterwards.

3. Every CLI channel held

At 6:00 AM ET today npm showed Claude Code stable at v2.1.287 with latest and next at v2.1.296 (published 12:58 PM ET Friday 9 Oct), Codex latest at v0.162.1 with alpha at v0.163.0-alpha.5 (8:48 PM ET Friday), Copilot CLI latest at v1.0.95 with prerelease at v1.0.96-2 (10:13 PM ET Friday), and Gemini CLI latest at v0.63.0 with preview at v0.64.0-preview.1. Those match the values in our last two briefs, on Saturday and Sunday. The only new build is Gemini CLI nightly v0.65.0-nightly.20261012, which reached npm at 9:32 PM ET on Sunday 11 Oct, and its GitHub release points at the same commit as the previous nightly, which reached npm at 9:42 PM ET on Friday 9 Oct. Gemini CLI v0.63.0 itself was released on GitHub at 4:38 PM ET on Tuesday 6 Oct. Cursor's main changelog still leads with Remote Control for local agents, and the Cursor CLI changelog still ends at v2026.10.08.

The Claude Code v2.1.296 subagent and workflow settings are covered in "Read the Large File Once. Compact the Subagent Early. Stable Stepped to v2.1.287.", and the gap between stable and the newer deny fixes is covered in "Channels Held. Stable Has Mods, Not the Deny Fixes."

Practical dev impact: Nothing on a command line needs a new pin this morning, so any version policy you set last week still stands. We would skip the Gemini CLI nightly unless you are testing something specific, since it carries no new commit.

Practical Impact Analysis

The useful work this Monday sits in editor settings and policy files rather than in the terminal, because the CLIs held still while GitHub shipped two controls that decide where an agent can act. In our reading the two are complementary: the JetBrains MCP startup switch limits which tools exist in a session to begin with, while local sandboxing limits what a command can touch once it runs. The managed default model adds a cost and approval decision alongside them, and it is the one control that belongs only to administrators.

Teams that standardize on JetBrains should treat the IDE version cutoff as the first task, since a developer who cannot update the plugin misses every new control in it. After that, we think it is worth agreeing on one written default for MCP startup and one for sandboxing, so that a freshly set up laptop matches policy before the first long agent session of the week. Teams that read our Wednesday brief and held off on sandboxing should revisit that call now that the feature is generally available.

Tutorial

We have not run this script end to end on every platform, so run it before you approve any update today. It prints each tracked channel and the local version, so you can compare them with Story 3. Install lines are commented out and labeled by channel, so nothing changes unless you uncomment one on purpose.

#!/usr/bin/env bash
# 1. Channel pointers. At 6:00 AM ET Mon 12 Oct npm showed the values in Story 3.
for p in @anthropic-ai/claude-code @openai/codex @github/copilot @google/gemini-cli; do
  echo "== $p"
  npm view "$p" dist-tags --json
done

# 2. What this machine runs
claude --version
codex --version
copilot --version
gemini --version

# 3. Claude Code stable channel, npm installs only, npm stable at 6:00 AM ET Mon 12 Oct:
# npm install -g @anthropic-ai/claude-code@2.1.287

# 4. Copilot CLI latest channel, npm installs only, npm latest at 6:00 AM ET Mon 12 Oct:
# npm install -g @github/copilot@1.0.95

If any npm view output differs from Story 3, something shipped after 6:00 AM ET, so read that release's notes before you update. Native installs follow their own pointers, so do not mix npm and native installs of the same tool on one machine. For JetBrains, confirm the IDE is v2025.2 or later, then open the Copilot plugin settings and find the MCP option for automatic server startup by its label rather than by any key name. Restart open sessions after any update and trust each tool's --version output over the tag.

Recommended AI prompt

Copy this paragraph into ChatGPT, Claude, Gemini, Grok, or whatever AI you use.