Streaming Markdown, session history, diffs and Manual, Accept Edits, Plan and Auto modes.
Claude Code. Inside Visual Studio.
A native Visual Studio integration for Claude Code: chat panel, planning, diffs, code context, IDE automation, debugging and Remote Control. Claude Code remains the agent; the extension gives it access to the real Visual Studio environment.
What problem does it solve?
Claude Code runs in a terminal, but Visual Studio knows things a terminal cannot see: the native diff viewer, Error List, projects and solution structure, the active document, debugger state, call stack, locals and the running application.
This extension connects Claude Code to that environment. Instead of copying code and diagnostics into a terminal, the agent uses IDE context and works on the real project.
Claude Code runs outside the Visual Studio process; the extension exposes IDE features to it through the Agent Client Protocol and IDE-control tools.
What do you get inside Visual Studio?
An implementation plan opens as a document. You can approve it or send it back with comments before implementation starts.
Every file changed by the session can be opened, accepted or rejected individually or in bulk.
The active document, multiple documents, selections and images can be attached without manually copying content.
Open and activate documents, add files to projects, work with the solution, build/rebuild/clean, read Error List and Output, and run DTE commands.
Breakpoints, start debugging, wait for a break, call stack, locals, expression evaluation, stepping, continue and stop.
While debugging, the agent can use UI Automation to inspect windows, click buttons, toggle checkboxes, type text and capture screenshots.
Drive the same session from claude.ai/code when Remote Control is enabled.
Workspace sessions, conversation title, elapsed time, token usage, activity and context state are visible in the panel.
When Visual Studio is in the background, notifications can signal completion, permission requests or a plan waiting for approval.
What does the workflow look like?
Open the solution and launch Claude Code from the Visual Studio panel. Add context when needed and describe the task as you would in a normal Claude Code session.
In Plan mode, you get the plan as a document first and can change it before the agent edits anything. File changes then appear in Changed Files and the diff view.
After editing, the agent can build the solution, read real diagnostics from Error List, start the debugger and inspect runtime state. So the work does not stop at a generated patch: the agent can compile, find the cause of a failure and fix it again.
Permissions and security.
Permission requests appear in the IDE and wait for your decision. In Manual mode the agent asks before every action that needs approval; Accept Edits and Auto ask less often.
The extension reuses the existing Claude Code login. It does not create a separate Visual Studio sign-in flow or take over credentials inside the extension process.
A session can send source code, unsaved edits, file paths, images and tool results to the configured model provider. Keep that in mind in repositories with confidential code.
Requirements and limitations.
The extension requires Visual Studio 2022 or 2026, the .NET 8 Runtime component in the Visual Studio installation, Node.js 22+ and a signed-in Claude Code CLI.
Capabilities depend on the ACP adapter and the features it advertises. The extension does not guarantee full parity with every version of the official Claude Code integrations.
Building from source.
Full VSIX builds and packaging require Windows and Visual Studio with the Visual Studio extension development workload.
F5 launches a separate Visual Studio Experimental Instance. Building the extension does not update your regular Visual Studio installation; install the rebuilt VSIX for everyday use.