IDE Integration

The terminal session is strongest for autonomous work. The IDE is strongest for inspection, local navigation, and human review. Use both deliberately.

Claude Code ships in two surfaces that share the same conversation history and ~/.claude/settings.json:

• A VS Code extension (and forks like Cursor, Windsurf, and Kiro) that provides a native graphical panel. This is the recommended way to use Claude Code in VS Code.
• A JetBrains plugin for IntelliJ IDEA, PyCharm, WebStorm, PhpStorm, GoLand, Android Studio, and most other JetBrains IDEs.

Either way, the underlying integration is the same: when Claude Code runs in (or is connected to) an IDE, it can open diffs in the editor's native diff viewer, read your current selection for @-mentions, and pull live diagnostics from the language server.

Effective Split

• Let Claude Code inspect, edit, run commands, and report verification.
• Use the IDE for reading dense diffs, checking UI copy, and resolving subtle merge conflicts.
• Keep terminal prompts focused on behavior, not on editor mechanics.

Connecting the IDE

There are two connection paths, and they behave differently.

For the JetBrains plugin, install it from the JetBrains Marketplace and restart the IDE, then run claude from the integrated terminal (or /ide from an external one). For VS Code, install the Claude Code extension (anthropic.claude-code) from the Extensions view or the Open VSX registry. The extension requires VS Code 1.98.0 or higher and includes the CLI, which you can reach from the integrated terminal.

Editor Context

When using an editor integration, make the selected context intentional. Passing too many open files can distract the agent from the actual task.

While connected, the CLI automatically includes your current editor selection and the path of the active file as context on each prompt. The transcript shows a ⧉ Selected N lines from <file> line when this happens. To insert an explicit @-mention reference instead, use the file-reference shortcut:

• VS Code: Option+K (Mac) / Alt+K (Windows/Linux) inserts something like @app.ts#5-10.
• JetBrains: Cmd+Option+K (Mac) / Alt+Ctrl+K (Windows/Linux) inserts something like @src/auth.ts#L1-99.

Useful context:

• The component being edited.
• The test that describes expected behavior.
• The route or API boundary affected by the change.

Noisy context:

• Large generated files.
• Unrelated tabs.
• Old scratch buffers.

Review Loop

After implementation, ask for a narrow review:

Review only the route and locale changes in this diff.
Prioritize bugs, unsupported fallbacks, and missing tests.
Return findings with file references.

When Claude proposes an edit, the integration shows a side-by-side diff in the editor's native viewer and asks for permission before writing. You can accept, reject, or edit the proposed content directly in the diff before accepting — if you do, Claude is told you modified it so it does not assume the file matches its original proposal. Use Plan mode when you want Claude to describe the change and wait for approval before touching anything; in the VS Code panel the plan opens as a markdown document you can comment on inline.

Set whether diffs open in the IDE or stay in the terminal with the diff tool setting: run /config and set the diff tool to auto (show in the IDE) or terminal (keep in the terminal).

The built-in IDE MCP server

When the VS Code extension is active, it runs a local MCP server named ide that the CLI connects to automatically. This is what lets the CLI open diffs in the native diff viewer, read your selection for @-mentions, and — in a Jupyter notebook — execute cells. The server binds to 127.0.0.1 on a random high port (so it is not reachable from other machines), and each activation generates a fresh auth token written to a lock file under ~/.claude/ide/ with 0600 permissions inside a 0700 directory.

The server is hidden from /mcp because there is nothing to configure, but it matters if your organization uses a PreToolUse hook to allowlist MCP tools. Most of its handlers are internal RPC the CLI uses for its own UI and are filtered out before the tool list reaches the model. Only two tools are visible to Claude:

Practical Tips

• Keep one terminal session per repository or worktree. Use claude --worktree <branch> (-w) to start Claude in an isolated git worktree with its own files and branch.
• Name branches by intent.
• Use editor search for local symbol navigation and Claude Code search for cross-repo discovery.
• Use Cmd+Esc / Ctrl+Esc to toggle focus between the editor and the Claude panel (VS Code), or to launch Claude Code from the editor (JetBrains).

• Reference terminal output in the VS Code extension with @terminal:name, where name is the terminal's title, so Claude can see logs and errors without copy-pasting.
• Let browser verification capture UI regressions rather than relying on visual guesses. The VS Code extension can drive Chrome via @browser when the Claude in Chrome extension is installed.

Version History