Configure options below — the preview updates in real time.

General & Enterprise Policies

Base application behavior, network connections, and enterprise management limits.

Disable HTTP/2Disables HTTP/2 for all requests, forcing HTTP/1.1. Used as a fallback for strict corporate proxies and VPNs.

Controls automatic update behavior for the IDE.

Enterprise policy that restricts which extensions can be installed. Leave empty to allow all.

Controls which Cursor Team IDs are allowed to log in. Users with unauthorized team IDs are forcefully logged out.

Configures a custom extension marketplace URL for enterprise deployments.

Determines which interface opens when accessing settings by default.

Chat & AI Assistant

Configurations for Cursor's natural language sidebar assistant.

Large ContextWhen enabled, chat utilizes much longer context windows. Note: uses more fast requests.
Collapse Input Box PillsCollapses file/context pills inside the chat pane or editor input box to save UI space.
Iterate on LintsIf enabled, the chat agent will automatically iterate on linter errors to attempt fixes without manual prompting.
Persist Chat HistoryPersist conversational chat history across workspace reloads and app restarts.
Enable Chat HistoryEnables indexing and tracking of local AI chat history.

Composer & YOLO Mode (Auto-Run)

Configurations for agentic loops where Cursor executes terminal commands autonomously.

Persist Composer HistoryPersists history for Composer workflows (multi-file generation).
YOLO Mode (Auto-Run)Allows the AI agent to write code, create files, and execute terminal commands continuously without waiting for explicit user approval.

The system instruction applied when YOLO mode is on (e.g., instructing the AI that "tests and build commands are always permitted").

Sandbox & Protection Settings

Constraints placed on Cursor's Agent to prevent destructive execution.

Choose how Agent runs execution tools. "Run in Sandbox" executes in an isolated environment, "Ask Every Time" prompts for confirmation, and "Run Everything" executes without confirmation.

Controls network access for sandboxed commands. Determines which domains the sandbox can reach.

Terminal commands the agent is permitted to run automatically outside of the sandbox without requiring user confirmation.

MCP (Model Context Protocol) tools that the agent is allowed to execute automatically outside of sandbox limits.

Browser ProtectionPrevents the agent from automatically running Chromium/Browser interaction tools without permission.
File Deletion ProtectionPrevents the agent from utilizing the rm command or deleting workspace files automatically.
Dotfile ProtectionPrevents the agent from automatically altering dotfiles (e.g., .gitignore, .env).
External File ProtectionPrevents the agent from creating or editing files outside the current project workspace directory.

Cursor Tab (Autocomplete)

Configurations for the inline Tab autocomplete engine.

Enable Partial AcceptsEnables partial accepts (Cmd/Ctrl + Right Arrow), letting you accept only the next word/chunk of a suggestion.
Suggestions in CommentsToggles AI suggestions inside comment lines. Disabling prevents suggestions while writing documentation.
Show Whitespace-Only ChangesDetermines if inline suggestions should be shown when the AI's generation only alters spacing or indentation.
Auto ImportAutomatically adds necessary import statements when accepting suggestions that reference unimported symbols (primarily TypeScript/JavaScript).

Language identifiers for which Cursor Tab suggestions should be disabled (e.g., scminput, yaml, markdown).

Terminal Preview BoxDetermines whether command suggestions in the terminal are rendered in an interactive preview box.

MCP Servers Configuration

Model Context Protocol server definitions. Configured via mcpServers in .cursor/mcp.json.

A key-value dictionary where each key is a tool name (e.g., github, database) and the value is a configuration object with command, args, and env properties.

AI Rules Configuration

Project-level rules for .cursor/rules/*.mdc files. The legacy .cursorrules file is deprecated; use Project Rules for better flexibility.

Contextual trigger description. The agent uses this semantically to decide when to load this rule into context. Required for intelligent application.

Defines the specific file patterns the agent must observe to trigger the rule.

*
Always ApplyIf true, bypasses the semantic routing and injects the rule into every single prompt in the workspace.

The actual prompt instructions and code standards (Markdown).