💻AI Codingintermediate

Cursor 0.42 Setup With Claude, A Working Indian Operator's Walkthrough

Install, model picker, $20/mo Pro, and the two settings that made it usable for me

Aditya Sharma··7 min read
Cursor editor with Claude model selected

import APIPriceLive from "@/components/data/APIPriceLive";

Cursor 0.42 ships a refreshed model picker, a sharper agent panel, and a rules layer that finally feels production-grade. I rolled it out as my main editor on the ThinkCentre running Ubuntu and on the MacBook Air for a week, billed against the Cursor Pro plan at $20/mo. This is the install that worked, and the two settings I would change on day one if I were starting over.

What you'll build

Cursor 0.42 set up on Linux or Mac, signed into Pro, configured to default to Claude Sonnet 4.6 for chat and inline-edit, with a .cursorrules file that scopes the agent to your stack. Roughly 20 minutes including the project rules pass.

Cursor 0.42 with Claude on my ThinkCentre Caption: Cursor 0.42 agent panel with Claude Sonnet 4.6 active.

Prerequisites

  • Linux x86_64 or Mac (Apple Silicon or Intel). I tested on Ubuntu 24.04 and macOS Sonoma.
  • 8GB RAM minimum, 16GB comfortable
  • Cursor Pro at $20/mo for Claude Sonnet 4.6 access. The free tier limits you to a smaller model
  • A working git project to test against

The free tier is enough to evaluate; the actual workflow with Claude needs Pro.

Step 1, install Cursor

On Linux, grab the AppImage from cursor.com and chmod it. On Mac, the dmg is signed and stock-friendly:

# Linux
wget https://downloads.cursor.com/production/cursor-0.42.0-x86_64.AppImage
chmod +x cursor-0.42.0-x86_64.AppImage
./cursor-0.42.0-x86_64.AppImage
# Mac (homebrew route)
brew install --cask cursor

Cursor first launch screen

The first launch asks you to import settings from VS Code. I imported keybindings only, skipped extensions. Importing every extension drags in heavy ones (Docker, GitLens) that you do not need for the AI flow.

Step 2, sign in to Pro

Use the in-app Sign In button. Cursor opens a browser to cursor.com, authenticates against your account, and writes a token back. You will see your plan label flip to Pro in the bottom-right.

Pro sign-in confirmed

If your account is on the free plan and you upgrade later, restart Cursor for it to pick up the new entitlement. I found the in-app refresh did not always pick up the change without a restart.

Step 3, set Claude Sonnet 4.6 as default

Open Settings (Ctrl+,), navigate to Cursor → Models. Toggle off any models you do not want in the picker (I keep Claude Sonnet 4.6, Claude Opus 4.6, GPT-5, and Gemini 3 Pro; everything else off so the picker is short).

Model picker with Claude Sonnet 4.6 default

In the agent dropdown at the top of the chat panel, pin Claude Sonnet 4.6 as default. Inline-edit (Ctrl+K) uses a separate model setting, set that one to Sonnet 4.6 too. Do not use Opus 4.6 by default; for inline-edit it is overkill and the latency cost shows.

Step 4, write a .cursorrules file

This is the single biggest quality lever in Cursor. Without it, the agent will introduce React Query when you use SWR, suggest Prisma when your stack is PocketBase, and try to install Tailwind v4 when your repo is locked to v3. Mine for an AutoKaam-style project:

# Stack
- Next.js 16 with static export, plain <img> not next/image
- PocketBase backend at /api/pb
- Tailwind v3, no JIT-mode arbitrary values
- TypeScript strict mode

# Don't
- Do not introduce a new package without asking
- Do not use next/image (breaks static export)
- Do not use console.log; use the logger from src/lib/logger.ts

# Voice
- File comments are short. One line max.
- Variable names are descriptive but not verbose.

cursorrules file in editor

Save as .cursorrules in the project root. Cursor picks it up at the next agent invocation.

Step 5, run a real task

The cleanest way to test the install is a real refactor:

You are working on src/app/page.tsx. Extract the data-fetching logic to a server-side helper at src/lib/news-loader.ts. Keep the page component pure. Run npm run typecheck after.

Cursor agent executing refactor

Cursor 0.42 surfaces a diff view before applying changes. Approve, reject, or edit each hunk inline.

First run

A typical session in 0.42 once configured:

You: refactor src/components/NewsCard.tsx, the image rendering logic should move to a separate ImageBox component

Cursor: I'll create src/components/ImageBox.tsx, extract the image and caption logic, update NewsCard to import it. Proceed?

You: yes

Cursor: [creates ImageBox.tsx, updates NewsCard.tsx]
       [runs typecheck]
       Done. tsc --noEmit clean.

Cursor session result

The diff view at the right shows exactly what changed. The agent does not commit, you stage and commit yourself.

What broke for me

Two real issues. First, on the AppImage on Ubuntu 24.04, sandbox permissions blocked Cursor from spawning child processes for typecheck. The agent would propose changes, hit a wall on tsc --noEmit, and report a vague timeout. The fix was running with --no-sandbox once: ./cursor-0.42.0.AppImage --no-sandbox. After the first launch, it remembered. On a stricter security posture you would package this differently; on a personal box it is fine.

Second, the Cursor agent would silently fail to read files larger than 1MB in my repo. The chat panel never said why. I cleared the indexer cache (rm -rf ~/.config/Cursor/User/globalStorage/<workspace>), restarted, and re-indexed. After that the large files (mostly generated TypeScript bundles) were readable again. The cause was a stuck index, not a setting.

What it costs

Item Cost
Cursor free tier $0 (limited)
Cursor Pro $20/mo (~Rs 1,660)
Cursor Business $40/user/mo
Claude API key (alt) Bring-your-own, pay per use

Pro covers Claude Sonnet 4.6 for chat plus inline-edit at fair-use limits. I hit the limits maybe twice in a normal week. If you are at the keyboard 8+ hours a day, Business gives you higher headroom and team rules.

When NOT to use this

Skip Cursor if your work is mostly in a remote terminal over SSH, you live in tmux/neovim, or you need an agent for multi-file refactors more often than inline edits. Claude Code on the CLI is a better fit for that. Cursor's strength is the in-IDE feedback loop with diffs, not headless agentic work.

If you cannot afford Pro, the free tier is genuinely usable for solo learners; the model is weaker but the workflow is the same. Upgrade when the model wall starts to bite.

Indian operator angle

Cursor bills in USD against your card. There is no INR plan, no Razorpay option. GST is not on the invoice; reverse-charge applies if you are claiming the expense in a registered business. My HDFC Regalia takes the foreign transaction at a 1.5% markup. Forex Markup eats Rs 25-30/mo, not material at the Pro tier.

For Indian-language coding work, Sonnet 4.6 inside Cursor handles Hindi commit messages and Hindi-named variables fine. Krutrim or Sarvam are not yet integrated as Cursor model options, so for the moment Claude is the path. The Cursor team has not announced an India-specific tier; I would not wait for one.

Related