GitHub’s September 22, 2026 changelog describes Copilot for JetBrains 1.18.0 as a broad update to agent-assisted development. The release adds public-preview assisted approvals, message re-editing with conversation and file rollback, Codex plan mode, expanded organization and enterprise customization support, and more control over the built-in GitHub MCP Server.
For developers, the practical change is greater control over how an agent acts inside a JetBrains IDE. For engineering and platform teams, the more important questions are how automatic approvals behave, which organization policies affect MCP access, and how shared instructions and skills fit into existing workflows.
What changed in Copilot for JetBrains 1.18.0
Assisted approvals for agent tool calls
Copilot agent sessions can now use AI-assisted approvals for tool calls. GitHub describes the feature as being in public preview. Calls categorized by the product as low risk can be approved automatically, while higher-risk actions continue to prompt the user.
The supplied announcement does not document the criteria used to classify a call as low or high risk. It also does not establish whether administrators can configure those classifications. Teams should therefore treat assisted approvals as a workflow convenience that requires review, not as a substitute for understanding what tools an agent can invoke.
Re-editing earlier messages
Users can re-edit an earlier message in a Copilot agent session. Before the replacement message is sent, Copilot rewinds the conversation and the associated file changes. This gives developers a way to correct a request without continuing from an unwanted instruction or set of edits.
The behavior is particularly relevant when an initial prompt was incomplete or sent before the intended scope was clear. It also means developers should understand that revising a message changes both the conversational history and the file state associated with that point in the session.
Codex plan mode
Codex agent sessions add a plan mode that lets users review, refine, or approve a plan before implementation begins. That creates a deliberate checkpoint between describing a task and allowing the agent to carry it out.
Plan review can be useful for checking scope and sequencing before implementation, but the announcement does not describe plan validation or guarantee that implementation will remain within the user’s intended boundaries. Developers should continue to inspect the proposed work and resulting changes.
Organization and enterprise customization
The update adds support for organization and enterprise skills and organization-managed custom instructions in local and Copilot agent sessions. These capabilities are intended to help apply shared development guidance alongside individual workflows.
There is an important availability qualification. GitHub’s feature matrix and customization reference identify several related JetBrains customization capabilities, including agent skills and custom instructions, as preview. The supplied documentation confirms that JetBrains has an Agent Customizations editor for managing supported custom agents, skills, workspace or personal instructions, and prompt files, but it does not fully document organization and enterprise inheritance or administration behavior.
MCP controls become more explicit
Copilot for JetBrains 1.18.0 adds controls for the built-in GitHub MCP Server. Users can enable or disable that built-in server without changing manually configured MCP servers. Agent sessions also provide persistent per-tool controls, making it possible to manage individual tools over time rather than treating the server as an all-or-nothing choice.
The built-in GitHub MCP Server remains enabled by default according to the changelog. GitHub’s MCP documentation says the server is maintained by GitHub and is available to GitHub users regardless of plan type. However, individual tools inherit the access requirements of their corresponding GitHub features, and users may be asked for additional permissions during use.
Organization and enterprise users need to account for policy controls as well. The supplied documentation says that members of organizations or enterprises using Copilot Business or Copilot Enterprise need the MCP servers in Copilot policy enabled to use MCP with Copilot. The documentation confirms the policy requirement but does not provide all of the JetBrains-specific setting names or configuration steps introduced by this update.
Interface, reliability, and environment changes
The release also includes a side-by-side chat panel switcher, browsable usage tips, a simplified welcome screen, a direct feedback link, improved MCP labeling and settings access, and restored shortcuts.
GitHub lists reliability improvements for inline chat, Codex startup, and multi-project windows. Embedded editors and message re-editing are restored on IntelliJ 2026.3 EAP builds, although the specific affected EAP build numbers are not identified in the supplied material.
Inline chat and its entry points are hidden in JetBrains Gateway and remote development environments. That changes how users in those environments access Copilot functionality and should be considered when standardizing development workflows across local and remote IDE use.
Users running JetBrains IDE version 2025.1 receive advance notice to upgrade to 2026.1 or later. GitHub states that support remains unchanged in this release, so the notice is an upgrade recommendation rather than a documented immediate removal of support.
What developers and administrators should do
- Review assisted-approval behavior. Before relying on automatic approvals, identify how agent sessions are used in your projects and pay attention to which calls still require user confirmation. The product’s risk-classification rules are not documented in the supplied sources.
- Use plan mode as a review checkpoint. For Codex tasks, inspect and refine the plan before approving implementation. Treat the plan as an opportunity to check scope rather than as a guarantee about the final changes.
- Test message re-editing on representative work. Confirm how rewinding affects the conversation and file changes in your workflow before using it for larger tasks.
- Audit MCP access. Review the built-in GitHub MCP Server setting and persistent per-tool controls. Verify the GitHub permissions required by the tools your team intends to use, and check whether the organization or enterprise MCP policy is enabled.
- Validate shared customization behavior. Use the JetBrains Agent Customizations editor to review supported skills and instructions. Organizations should test how managed guidance appears alongside workspace and personal customizations because the full inheritance model is not documented here.
- Plan IDE upgrades where appropriate. Teams still using JetBrains IDE 2025.1 should evaluate the announced path to 2026.1 or later, while accounting for the fact that the release states support remains unchanged.
Availability and remaining limits
GitHub identifies the release as Copilot for JetBrains 1.18.0, but the supplied evidence does not independently confirm exact JetBrains Marketplace metadata or availability for that plugin version. Assisted approvals are explicitly in public preview. No pricing, geographic rollout, or rollout percentage is provided.
The MCP controls provide more user control over tool availability, but they should not be interpreted as unrestricted access or as a formal security guarantee. Feature permissions, additional approval requests, and organization policies can still affect whether a tool is usable. Similarly, the announcement’s organization and enterprise customization support should be read alongside the documentation’s preview labels and the unresolved details about inheritance and administration.
For teams adopting the update, the main operational change is not simply more agent functionality. It is the addition of several review and control points: automatic approval for some tool calls, plan approval before Codex implementation, reversible message editing, and persistent MCP tool controls. Those controls are most useful when teams test them against their own permissions, IDE environments, and development policies.
