GitHub released GitHub Copilot for JetBrains 1.18.0 on September 22, 2026, with changes to agent workflows, tool approvals, organizational customization, planning with the Codex agent, MCP controls, and user experience. To know whether a new capability is available in your environment, however, looking only at the version number is not enough. You need to separate plugin compatibility with the IDE, feature support in JetBrains, and any preview or organization administration requirements.
What arrived in GitHub Copilot for JetBrains 1.18.0
The official changelog highlights six main areas. Assisted approvals are in public preview for Copilot agent sessions and can automatically approve tool calls considered low risk, while higher-risk actions still ask the user for a decision. It is also now possible to re-edit an earlier message in an agent session: before the replacement is sent, Copilot rewinds the conversation and file changes to that point.
Local and Copilot agent sessions now accept organization and enterprise skills, as well as organization-managed custom instructions. The Codex agent gained plan mode, where a developer can review, refine, or approve a plan before implementation.
For Model Context Protocol integration, a new setting lets you turn the built-in GitHub MCP Server on or off without changing manually configured MCP servers. The built-in server remains enabled by default, and Copilot agent sessions now provide persistent per-tool controls for MCP servers.
The release also adds a panel switcher to keep chat alongside the session list, browsable tips above the chat input, and improvements to settings discoverability. For fixes, GitHub cites better inline chat reliability, respect for the selected thinking effort and context window, Codex session startup fixes, and improvements across multiple project windows.
Where to verify plugin compatibility with your JetBrains IDE
For product and version compatibility, the practical reference is the GitHub Copilot versions page in JetBrains Marketplace. Each published package shows its channel, date, compatibility range, and accepted JetBrains products. This matters because the same functional Copilot release can have different packages for different IntelliJ Platform lines and operating systems.
GitHub's installation documentation complements this check with the general list of supported JetBrains IDEs and instructions to install or update the plugin through the Marketplace. Listed environments include IntelliJ IDEA, Android Studio, CLion, DataGrip, DataSpell, GoLand, PhpStorm, PyCharm, Rider, RubyMine, RustRover, and WebStorm, along with other JetBrains clients named in the documentation.
In practice, check your exact IDE version first and then open the corresponding plugin entry in the Marketplace. If a compatible package does not appear in the IDE, verify the build range and product declared on the version page instead of assuming that every changelog feature applies to every JetBrains version.
Where to check which features actually work in JetBrains
The GitHub Copilot feature matrix is the best reference for separating basic plugin compatibility from feature availability. It compares features across IDEs and also across recent extension versions. In the current table, JetBrains is shown with support for capabilities such as agent mode, chat, checkpoints, code completion, code referencing, Copilot code review, edit mode, MCP, and workspace indexing. Other items, including agent skills, custom agents, custom instructions, next edit suggestions, prompt files, and vision, may be marked as preview.
GitHub itself warns that this matrix is in public preview and can change. For a reliable check, use the matrix together with the changelog for the installed version. The changelog says what changed in that release; the matrix shows the support status of the feature across the IDE ecosystem.
Two important exceptions in this update
Two notes help avoid incorrect assumptions. First, inline chat and its entry points are now hidden in JetBrains Gateway and remote development environments. An environment can therefore remain part of the JetBrains ecosystem while not exposing the same local user surface.
Second, users of JetBrains IDE version 2025.1 will start seeing advance notice to upgrade to 2026.1 or later. GitHub says support is unchanged in this release, so this is a transition notice and does not mean that 1.18.0 immediately dropped support for the 2025.1 line.
How to verify before looking for a feature
- Check the IDE and build: note the JetBrains product and installed version.
- Check the plugin: open Plugins settings and identify the installed GitHub Copilot version.
- Validate in the Marketplace: open the versions page and confirm the compatibility range for the corresponding package.
- Validate the feature: consult the Copilot feature matrix to see whether the capability is supported, in preview, or unavailable in JetBrains.
- Read the version changelog: confirm specific conditions, behavior changes, and release exceptions.
- Consider organization policies: managed features, shared skills, instructions, and integrations can depend on administrator-defined settings in addition to technical compatibility.
What changes for users of agents and MCP
Version 1.18.0 concentrates many of its changes on agentic workflows. Assisted approvals reduce interruptions for low-risk calls, but they do not remove the user's decision for higher-risk actions. Message re-editing provides a way to return to an earlier point in a session, including the state of file changes, before redirecting the task.
For teams, organization-managed skills and instructions can standardize behavior across Local and Copilot agent sessions. Codex plan mode adds an explicit plan review step before implementation. For MCP, the more granular controls help separate activation of the built-in GitHub server from manual configurations and persist per-tool controls.
These changes do not mean that every feature appears identically in every JetBrains product or plugin version. The safest verification method is to combine the Marketplace, feature matrix, and changelog because each source answers a different question.
