Why Cursor over Claude Code?
- Code-generation diffs are easier to review and roll back (the checkout system is great).
- You can test and use other models for different tasks (
o3
for complex tasks,gemini-2.5-pro
for creativity,grok-code-fast-1
for quick tasks, etc.).* - Good IDE integration (e.g., accept/reject actions work great across the UI).
*That said, I’m only using claude-4.5-sonnet
since its launch.
How to 10x Productivity on Cursor
Good pre-prompt
Use .cursorrules
or similar to enforce behavior. For example:
// .cursorrules for a frontend project
- Always use the project’s Tailwind colors.
- Always use named exports with object parameters: `sum({ a, b })` instead of `sum(a, b)`.
- Use "pnpm" instead of "npm" for all package-management commands.
- Use the "use client" directive when using client-side state like useState, useEffect, etc.
- Don’t modify package.json dependencies directly.
- Always write fail-safe code using optional chaining and default values.
- Use DayJS for all date and time operations — don’t use `new Date()`, `Date.now()`, or other date functions.
- When searching the codebase for context, do at least three semantic searches and three grep searches (do more if needed).
// .cursorrules for a blockchain dapp project
- Don’t run dev-server commands; let me do it. If you want to run something, ask me first.
- You won’t be able to run smart-contract deployment commands when needed; ask me and I will do it.
- Run `npm run typecheck` after making changes in the frontend and fix errors. Avoid `any` types.
- Run `npm run compile:check` after making changes to the contracts and fix errors.
- When updating a smart contract, remember to generate the ABI and update the FE accordingly.
- When coding a smart contract, check BYTECODE_OPTIMIZATION.md for best practices.
- Don’t write .md files unless asked to.
- If updating a smart contract, at the end ask me to deploy the contract with `npm run redeploy:all`.
Good MCPs
- Atlassian MCP (Jira ticket integration)
- Chrome DevTools MCP (or Cursor’s in-browser)
-
GitHub MCPI’d rather use GH CLI commands instead - Figma MCP
- Neon MCP (for hobby projects)
I’ve also heard these are good:
- Svelte MCP
- Stripe MCP
- Sentry MCP
- AWS / Cloudflare / Vercel MCPs
Enable them on demand to avoid filling up context.
Fireship has a good video showing several of them:
Automation
Create .md
files with reusable instructions to automate small wins, such as:
- Creating and enhancing GitHub PRs and releases with links and more (Atlassian MCP + GitHub CLI)
- UI testing (Chrome MCP / Playwright MCP / Cursor’s browser)
- SVG → React component conversion using a template
- API testing
Other 10x Modifiers
- Use the “Plan” feature — always try it first.
- Voice transcription tools to dictate detailed prompts.
- Screenshot and text-recognition tools to provide extra context.
Other Cursor features I don’t use much
- Bugbot (PR review — too expensive)
- cursor-agent (too specific tasks)
- @doc (usually web search does the trick)
- /commands (I kinda prefer .md files)
- Background Agents (It's quicker just to open the project locally)
Top comments (0)