Cursor and GitHub Copilot

The two most popular IDE-integrated AI coding assistants. Neither reaches Claude Code's autonomy level, but both integrate into existing workflows with less friction.

Cursor

A VS Code fork with AI built into the editor itself. Not a plugin. A separate application with native AI integration throughout the IDE.

Key Features

Composer / Agent mode: Multi-file editing with a goal. Describe a feature; Cursor reads relevant files, proposes a plan, makes edits across multiple files. Less autonomous than Claude Code (no shell execution by default), but more visual.

Inline editing (Cmd+K): Highlight code, describe a change. Cursor applies it.

Chat (Cmd+L): Context-aware chat that references the file you're editing, your entire codebase (via repo indexing), docs, and web search.

Rules (.cursorrules): The equivalent of CLAUDE.md for Cursor. Define project conventions, preferred patterns, what to avoid.

# .cursorrules
You are helping with a Python FastAPI project.
- Use async/await throughout
- Use Pydantic v2 for all data models
- No print statements — use structlog
- Tests use pytest with pytest-asyncio mode=auto

MCP support: Cursor supports MCP servers. Add to ~/.cursor/mcp.json or project-level .cursor/mcp.json.

Models

Cursor routes between models based on task type. You can select specific models:

• Claude Sonnet 4.6 — default for agent/composer tasks
• Claude Haiku 4.5 — fast inline completions
• GPT-4o — alternative for code
• Cursor's own Cursor-Fast for completions

Pricing

$20/month (Pro). Includes 500 fast requests and unlimited slow requests per month. Team plan at $40/user/month.

GitHub Copilot

Microsoft/GitHub's AI assistant. More conservative than Cursor. Inline suggestions first, chat second.

Features

Inline completions: Ghost text suggestions as you type. The original Copilot feature. Tab to accept.

Copilot Chat: Conversational chat within VS Code. Context-aware. Can reference open files, selected code, terminal output.

Copilot Edits: Multi-file editing via natural language (newer feature). Similar to Cursor Composer but less developed.

MCP support (VS Code 1.99+): Full MCP host support. Add MCP servers to settings.json:

{
  "github.copilot.mcp.servers": {
    "my-server": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "my_mcp_server"]
    }
  }
}

Models

Copilot uses Claude Sonnet 4.6, GPT-4o, and Gemini models depending on task. The model routing is partially transparent via the model picker in VS Code.

Pricing

Individual: $10/month. Business: $19/user/month. Enterprise: $39/user/month.

Comparison Matrix

Aider

Open-source CLI alternative to both. Repository-aware, multi-file editing, works with any LLM via API.

pip install aider-chat
aider --model claude-sonnet-4-6 --file src/app.py

Architect mode: A planning model (Opus) designs the change; an editor model (Sonnet/Haiku) implements it. Good for complex refactors.

Repo map: Aider builds a tree-sitter based map of the codebase for context-efficient references.

Choosing

• Claude Code — multi-step autonomous tasks, refactors across many files, anything requiring shell
• Cursor — daily development where you want an IDE-native experience with good multi-file editing
• Copilot — team standardisation, tight GitHub integration, or you prefer staying in your existing editor
• Aider — open-source preference, want to pick your own model, CLI-centric workflow

Key Facts

• Cursor pricing: $20/month Pro (500 fast requests + unlimited slow); $40/user/month Team
• GitHub Copilot pricing: $10/month Individual, $19/month Business, $39/month Enterprise
• Cursor uses VS Code fork architecture — not a plugin, a separate application
• Cursor MCP config: ~/.cursor/mcp.json or .cursor/mcp.json (project-level)
• Copilot MCP support added in VS Code 1.99+
• Cursor default agent model: Claude Sonnet 4.6; fast completions: Claude Haiku 4.5
• .cursorrules is the Cursor equivalent of CLAUDE.md
• Aider architect mode: Opus plans, Sonnet/Haiku implements

Common Failure Cases

.cursorrules instructions are silently ignored because the file is placed in a subdirectory rather than the repo root
Why: Cursor loads .cursorrules only from the workspace root (the directory opened as the project in Cursor); if the file is in a src/ subdirectory or a monorepo package, it is not found and no instructions are applied.
Detect: Cursor ignores project-specific rules (e.g., uses print despite a rule saying to use structlog); checking ~/.cursor/logs/ or the Composer context shows no rules file loaded.
Fix: place .cursorrules at the root of the folder opened as the Cursor workspace; for monorepos, create a root-level file that references the conventions for all packages.

Cursor Composer edits the wrong file because the repo index is stale and maps a function name to an outdated file location
Why: Cursor indexes the codebase at startup; if files were moved or renamed outside of Cursor (e.g., a git pull), the index may still reference old paths; Composer proposes edits to the stale location, which may be a deleted or renamed file.
Detect: Composer opens a file that doesn't match the current codebase structure; git status shows a new file created at the old path; the correct file at the new path is unchanged.
Fix: force a re-index by reloading the Cursor window (Cmd+Shift+P → "Reload Window"); or close and reopen the project after large structural changes.

GitHub Copilot suggests code that references an internal API that was deprecated in a recent version, because its training data cutoff predates the deprecation
Why: Copilot's suggestions are based on training data with a cutoff date; for rapidly evolving frameworks (LangChain, Next.js App Router, SQLAlchemy 2.0), Copilot may suggest patterns from older versions that are no longer valid.
Detect: Copilot suggests from langchain.chat_models import ChatOpenAI (deprecated) instead of from langchain_openai import ChatOpenAI; the suggested code runs but emits deprecation warnings or fails at runtime.
Fix: when working with rapidly evolving frameworks, use Cursor or Claude Code with MCP documentation context tools that can fetch current API docs; treat Copilot suggestions for recently changed APIs with extra skepticism.

Connections

• ai-tools/claude-code — Claude Code comparison; higher autonomy, shell execution, MCP ecosystem
• protocols/mcp — MCP servers available to Cursor, Copilot, and Claude Code alike
• ai-tools/aider — open-source CLI alternative; architect mode, repo map; lower autonomy than Claude Code but model-agnostic
• ai-tools/cline — VS Code agent with 30+ LLM providers and Plan/Act mode; direct competitor to Cursor Composer

Open Questions

• How does Cursor's repo index compare to Claude Code's full file-read approach for large monorepos?
• Will Copilot Edits eventually reach Cursor Composer parity in multi-file edit quality?
• What is the real-world cost comparison between Cursor's fixed $20/month and Claude Code's usage-based pricing for a typical developer?