AI Summary Hub

Cursor

AI-powered code editor and pair-programming tool.

Definition

Cursor is an AI-powered code editor forked from VS Code that embeds LLMs directly into every part of the editing experience. Unlike extensions bolted onto an existing editor, Cursor owns the entire editor surface, which allows it to build features such as multi-file diff previews, codebase-wide semantic search, and chat with full project context that are difficult to replicate through extension APIs alone.

The editor supports multiple model backends (Claude 3.5/3.7, GPT-4o, local models via Ollama) and allows users to define project-level instructions through .cursorrules files, steering the model's style, conventions, and tooling assumptions. Context is managed through an embedding-based codebase index that makes relevant files available to the model without manual selection, enabling workflows closer to pair programming than inline autocomplete.

Compared to GitHub Copilot, Cursor offers deeper project context, inline diff-based edits, and a full chat panel; compared to Claude Code, it is a GUI-first experience centered on the editor rather than the terminal. All three tools use LLMs for code generation, but differ in interface, context management, and agent depth.

How it works

Inline edit (Cmd+K)

Chat panel (Cmd+L)

Key features

Codebase index — embeds the repo for semantic search. Composer — multi-file agent-style edits. Tab completion — context-aware next-line and block completion. .cursorrules — persistent project instructions for the model. MCP support — tool use via the Model Context Protocol.

When to use / When NOT to use

ScenarioUse CursorDo NOT use Cursor
In-editor coding with full project contextYes — codebase indexing provides deep context
Multi-file refactoring with visual diffsYes — Composer and diff views
Pair programming with explanations in chatYes — persistent chat panel
Terminal-first or headless environmentsUse Claude Code CLI instead
IDE-agnostic completion in JetBrains or NeovimUse GitHub Copilot for broader IDE coverage
Lightweight extension on existing VS CodeCopilot or Codeium add-ons have less overhead

Comparisons

FeatureCursorGitHub CopilotClaude Code
Base interfaceFull VS Code forkIDE extensionTerminal + IDE extension
Project contextCodebase index (embeddings)Open files onlyFull repo via CLI
Multi-file editsYes (Composer)LimitedYes (terminal + IDE)
Agent capabilitiesComposer, MCPCopilot WorkspaceClaude agents
Model choiceMultiple (Claude, GPT-4o, local)OpenAI / GitHubClaude (Anthropic)
Rules / config.cursorrules fileNo project rulesCLAUDE.md
PricingSubscription (hobby free tier)SubscriptionSubscription (Pro+)

Pros and cons

ProsCons
Deep project context via codebase indexingRequires switching from existing VS Code setup
Supports multiple LLM backendsCodebase indexing may expose code to third-party servers
Project-level rules steer model behaviorHeavy resource usage compared to lightweight extensions
Visual diff previews make edits reviewableContext limits still apply; very large repos need selective inclusion

Code examples

// .cursorrules — project instructions for the model
{
  "rules": [
    "This is a TypeScript/React project using Tailwind CSS.",
    "Prefer functional components and hooks over class components.",
    "Always add JSDoc comments to exported functions.",
    "Use the existing `api/` client for all HTTP calls; do not use fetch directly.",
    "Tests are written with Vitest; always add a test for new utility functions."
  ]
}

Tips for effective use

  • Keep .cursorrules short and specific — long rules dilute the model's attention.
  • Use @file or @folder mentions in chat to pin relevant context.
  • For large repos, exclude generated files (node_modules, dist, .next) from the codebase index to reduce noise.
  • Accept Composer suggestions incrementally — review each diff before accepting the next change.
  • Pair Cursor with a linter and type checker so the model gets immediate feedback on generated code quality.

Practical resources

See also