CodeCursor (Cursor for VS Code)

Converts natural language prompts into executable code by routing requests through Cursor's server infrastructure to OpenAI GPT models, streaming generated code back to VS Code as a live text diff with accept/reject controls. The extension intercepts the generation stream and renders it incrementally in an inline notification panel, allowing users to preview changes before applying them to the document.

Opens a persistent chat panel in VS Code's sidebar that maintains conversation context about the currently open document or selected code. Messages are routed through Cursor's server to GPT models, enabling developers to ask questions about code semantics, request explanations, or discuss implementation details without leaving the editor. The chat maintains multi-turn conversation history within a session.

Automatically scopes all code generation and explanation requests to the currently open document, using the full file content as implicit context for prompts. The extension does not require users to manually specify file context — it's automatically included in every request. This enables context-aware generation without explicit context management, though it limits awareness to single-file scope.

Generates entire project directory structures and boilerplate code from natural language descriptions by routing requests to GPT models via Cursor's server. The extension creates files and folders in the current workspace, with warnings if the workspace is non-empty to prevent accidental overwrites. This feature is marked experimental and may have undefined behavior with concurrent generation requests.

Allows users to override the default Cursor server backend by providing custom OpenAI API keys in extension settings, enabling model selection and cost control. The extension routes all requests through the provided API key instead of Cursor's infrastructure, though the connection still flows through Cursor's server as an intermediary rather than direct client-to-OpenAI communication. Configuration is stored in VS Code's extension settings.

Enables users to select code snippets in the editor before triggering generation, automatically using the selection as context for code generation prompts. When code is generated, the selected text is replaced with the generated output in a single atomic operation, with the change shown as a diff in the notification panel before acceptance. This allows targeted code modification without affecting surrounding code.

Displays real-time progress indicators in VS Code's status bar during code generation and project scaffolding operations, allowing users to cancel in-progress requests by clicking the status bar item. The status bar shows operation type (generating code, creating project) and provides a clickable interface to abort requests or reopen completed results without re-running generation.

Routes all AI requests through Cursor's managed server infrastructure by default, which handles authentication, rate limiting, and model selection. If the Cursor server becomes unstable or unavailable, users can configure custom OpenAI API keys to bypass Cursor's infrastructure entirely. The extension abstracts away the routing logic, presenting a unified interface regardless of backend selection.

Exposes code generation and project creation features through VS Code's command palette with 'CodeCursor' prefixed commands, allowing keyboard-driven access without sidebar interaction. Users invoke commands like 'CodeCursor: Generate Code' or 'CodeCursor: Generate Project' from the command palette, which opens a prompt input dialog for natural language specifications. This enables rapid feature access for power users and keyboard-first workflows.

Provides a dedicated sidebar panel accessible via the CodeCursor icon in VS Code's activity bar, serving as the primary UI for conversational code explanation and navigation between features. The sidebar maintains chat history within a session and provides quick access to code generation and project creation features without using the command palette. The panel persists across editor focus changes, allowing continuous conversation while coding.

Implements a privacy-first design where the extension explicitly claims NOT to collect user code, environment data, or tracking information. Code sent to Cursor's server or custom OpenAI API is used only for the immediate generation request and not retained for training or analytics. This design choice is documented in the extension's security section, though implementation details of how code is handled server-side remain undocumented.