Choosing between Claude Code and Cursor for real product work

Claude Code vs Cursor for real product work: agent or editor, models and pricing as of November 2026, rules, skills, MCP, review, and Figma to code.

Summarize with
A laptop with lines of code appearing and a gear turning beside it

Claude Code vs Cursor comes down to how you want to work. Claude Code is an agent: you describe the job, it reads the codebase, edits files and runs commands, and you review the diff. It runs in a terminal, an IDE, a desktop app or the web, with Claude models only. Cursor is an editor: an IDE with inline completions, an agent panel, cloud agents and models from several companies. Since August 2026 it belongs to SpaceX and is sold in a bundle with Grok and X, so its pricing has changed. Pick Claude Code to hand off whole tasks. Pick Cursor to steer inside the editor. Many people run both.

I'm a design engineer: I design products and build them in production code. I build this site with Claude Code, so my notes on it are first-hand. I don't use Cursor, so everything about Cursor here comes from its own docs and pricing pages, read on 4 November 2026.

The short answer: agent or editor

The real difference is where your attention sits.

With Claude Code, your attention sits on the task and the diff. You say what you want, approve the commands it may run, and read the change when it's done. The terminal is the full version; the IDE extensions add inline diffs and plan review.

With Cursor, your attention sits on the file. Its docs describe Tab (inline completions), an Agent in the side pane, Projects where a coordinator agent splits work across agents, and cloud agents on Cursor's machines. A Cursor CLI, started with agent, has Agent, Plan and Ask modes.

Claude Code vs Cursor at a glance

Claude CodeCursor (from its docs)
Version I checked2.1.317, released 3 November 2026IDE and Agents Window 3.27, released 22 October 2026
Main way of workingAgent: terminal, IDE extension, desktop app, webEditor: IDE with Tab, side panel Agent, cloud agents, CLI
ModelsClaude only (Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1)Grok 4.7, 4.6, 4.5 and Composer 2.5 included; Claude, GPT and Gemini at API list price
Entry paid planClaude Pro, $20 a month billed monthly, $17 billed yearlyPlus, $8 a month (Cursor's own models only)
Plan with third-party modelsNot applicablePremium, $30 a month, plus usage
Project instructionsCLAUDE.md, .claude/rules, reads AGENTS.md if there's no CLAUDE.md.cursor/rules (.mdc files), User Rules, Team Rules, AGENTS.md
SkillsAgent Skills in .claude/skills and ~/.claude/skillsAgent Skills in .agents/skills, and it also loads .claude/skills
HooksFive hook types: command, http, mcp_tool, prompt, agenthooks.json scripts; can load Claude Code hooks
PR reviewCode Review on Team and Enterprise; /code-review locally on any planBugbot on Premium and above; Agent Review locally

Models: Claude only, or Grok plus metered extras

Claude Code runs Claude models. On the Anthropic API, its opus, sonnet and haiku aliases point to Opus 5.5, Sonnet 5.5 and Haiku 5.5, and fable to Fable 5.1. If you want GPT or Gemini, it isn't the tool.

Cursor's docs list models from OpenAI, Anthropic, Google and SpaceXAI. The catch is billing. Cursor's own models (Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5) draw from your weekly pool. Claude, GPT and Gemini don't; on Premium and above you pay for them from prepaid credit at API list price. Cursor lists Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, which matches Anthropic's own API price list. On the $8 Plus plan, third-party models aren't available at all.

Many people chose Cursor to run Claude on a flat plan. Today, Claude inside Cursor is metered like the API.

Pricing and limits, checked 4 November 2026

Claude Code comes with Claude plans. Free doesn't include it. Pro is $20 a month billed monthly, or $17 a month billed yearly ($200 up front). Max is $100 a month with 5 times Pro usage. Ultra is $200 a month with 10 times Pro usage each week and no 5 hour limit for verified accounts. Limits reset every 5 hours and every week. Team seats are $25 a month billed monthly or $20 billed yearly (standard), and $125 or $100 (premium, 5 times the usage), for 2 to 150 people. Enterprise is $20 a seat a month billed yearly, with a 20 seat minimum, plus usage at API rates.

On API billing, Anthropic's cost docs put the enterprise average at around $13 per developer per active day.

Cursor is now sold as part of The X Subscription, one subscription for Grok, Cursor and X. Free is $0 with limited use. Plus is $8 a month. Premium is $30 a month with 4 times Plus usage, and adds cloud agents, Bugbot and third-party models. Super is $100 (15 times Plus) and Ultra is $200 (50 times Plus). Yearly billing is $80, $300, $1,000 and $2,000. The usage pool refreshes weekly and is shared with the Grok apps. Teams seats cost $30, $100 or $200 a month, and on Teams and Enterprise, eligible third-party model requests also carry a Cursor Token Rate of $0.25 per million tokens. Subscribers on Cursor's older plans, such as Pro and Pro+, keep them as legacy plans.

Many comparisons on page one still quote Cursor's old $20 Pro plan. Check the pricing page on the day you buy.

Context and large codebases

Both tools search your code instead of loading all of it. Cursor's docs describe Instant Grep, a local index that uploads no file paths and stores no embeddings, and an Explore subagent that searches in its own context and returns a summary. In Claude Code, Sonnet 5.5 and recent Opus versions have a native 1 million token window on the Anthropic API, and long sessions compact automatically (/autocompact sets when). What matters more is your instruction files: clear house rules help a large repo more than a bigger window.

Rules, skills and hooks

  • Rules. Claude Code reads CLAUDE.md files, path-scoped rules in .claude/rules, and since version 2.1.277 an AGENTS.md when no CLAUDE.md exists. Cursor reads .cursor/rules files (they must be .mdc, a plain .md there is ignored), User Rules, Team Rules from its dashboard, and AGENTS.md. One AGENTS.md can serve both.
  • Skills. Both follow the Agent Skills format: a folder with a SKILL.md and optional scripts and references. Cursor's docs say it loads skills from .claude/skills and ~/.claude/skills for compatibility. My free Claude skills for design work, built on Jakub Krehel's MIT-licensed better-* skills, install into ~/.claude/skills for Claude Code; the install guide has the steps.
  • Hooks. Claude Code hooks run at points in the agent loop, such as before a tool runs, and can block an action; its docs list five hook types. Cursor hooks are hooks.json scripts with events such as beforeShellExecution and afterFileEdit, and it can load hooks from Claude Code.

MCP: same protocol, different defaults

Both connect to MCP servers over local stdio and remote HTTP. Claude Code adds one with claude mcp add or a shared .mcp.json; Cursor uses its marketplace or mcp.json. Cursor's security docs say each MCP tool call needs approval unless you allowlist it. Claude Code applies its permission modes and allow rules.

Review and safety

Claude Code has permission modes. Manual (the default) asks on the first use of each tool. acceptEdits accepts file edits. Plan explores without editing source. Auto skips routine prompts and checks shell and network actions with a background classifier. bypassPermissions skips prompts, and the docs say to use it only in containers or VMs. Code Review, on Team and Enterprise, runs several agents on each GitHub pull request and comments inline; /code-review works locally on any plan.

Cursor, per its docs, edits workspace files without approval (except configuration files) and saves to disk at once, so version control is your undo. Terminal commands need approval by default. Checkpoints roll files back within a session. Bugbot reviews pull requests on Premium and above.

If you plan to run both, know this: on 29 October 2026, Cursor reported that malicious copies of the Claude Code extension, versions 2.1.312 and 2.1.313, were in the Cursor VS Code Extension Marketplace for about 11 hours. Install extensions from the official listing and check the version.

Neither tool's review replaces a person. I wrote up the checks I run on generated UI in reviewing AI-generated UI code.

Team use

Claude's Team plan adds central billing, SSO and mixed seats; Enterprise adds SCIM, audit logs and admin spend limits. Cursor's Teams adds a team marketplace for rules and skills, Team Rules and SSO; Enterprise adds model and MCP access controls and audit logs. Either way, keep shared rules in the repo, not in personal settings.

Figma to code and UI work

This is where a design engineer feels the difference.

Both connect to Figma's official MCP server, which gives the agent variables, components and layout from a frame. Figma's guide installs it in Claude Code with claude plugin install figma@claude-plugins-official, and in Cursor with /add-plugin figma. Output depends on the file more than the tool: named layers, and variables and components that match your code.

Cursor has one feature built for UI work that I have no exact match for in my Claude Code setup. Design Mode, in the browser inside its Agents Window, lets you click an element on your running app, box an area or edit text in place, attach a note, and send it to the agent with a screenshot. It needs Cursor 3.24.18 or later for the Annotate button. For small visual fixes, pointing beats describing.

Claude Code is stronger, in my use, for a whole flow: a new page, a component and its states, copy wired into the CMS and a migration, reviewed as one diff. My full process is in the design to code workflow.

When to pick which

Pick Claude Code if you hand off tasks that cross many files, work in a terminal, VS Code or JetBrains, are happy with Claude models only, and want one flat plan.

Pick Cursor if you want inline completions while you type, a visual editor with checkpoints, a choice of model vendors in one place, or Design Mode for pointing at UI. Budget for third-party model usage if you want Claude or GPT inside it.

Use both if you like Cursor as your editor and Claude Code as the agent. Anthropic publishes the Claude Code extension for Cursor, and a shared AGENTS.md plus skills in .claude/skills keeps both tools on the same rules. Switching? Move your rules first (.cursor/rules into CLAUDE.md or AGENTS.md, or back), then skills, then MCP servers.

FAQ

Is Claude Code better than Cursor?

Neither wins outright. Vendor benchmarks measure their own setups, so try both on your own repo for a week.

Can I use Claude models in Cursor?

Yes, on Premium ($30 a month) and above, billed from prepaid credit at API list price, as of 4 November 2026. The $8 Plus plan includes only Cursor's own models.

Does Claude Code have a free plan?

No. Claude's Free plan doesn't include Claude Code. Pro, at $20 a month billed monthly, is the cheapest plan that does.

Will my Cursor rules work in Claude Code?

Not as they are. .mdc rules are Cursor's format. Move the content into CLAUDE.md, .claude/rules or an AGENTS.md, which Claude Code reads when there's no CLAUDE.md. Skills in the Agent Skills format work in both.

Which is better for a design engineer?

For pointing at a running page and fixing visuals, Cursor's Design Mode is the more direct tool. For building whole screens from Figma with tokens, states and checks in one reviewed change, I use Claude Code.

Sources and method

I compared the official docs and pricing pages of both tools and read the current top results for this query on 4 November 2026. I use Claude Code daily; Cursor is described only from its own documentation. Versions and prices change often, so check them before you buy.

  • Claude Code changelog (version 2.1.317), date read 4 November 2026: code.claude.com
  • Claude Code overview, date read 4 November 2026: code.claude.com
  • Claude Code model configuration, date read 4 November 2026: code.claude.com
  • Claude Code permissions, date read 4 November 2026: code.claude.com
  • Claude Code memory, CLAUDE.md and AGENTS.md, date read 4 November 2026: code.claude.com
  • Claude Code hooks, date read 4 November 2026: code.claude.com
  • Claude Code costs, date read 4 November 2026: code.claude.com
  • Claude Code Code Review, date read 4 November 2026: code.claude.com
  • Claude plans and pricing, date read 4 November 2026: claude.com
  • Anthropic API pricing, date read 4 November 2026: platform.claude.com
  • Cursor pricing, date read 4 November 2026: cursor.com
  • Cursor docs, models and pricing, date read 4 November 2026: cursor.com
  • Cursor release notes (3.27), date read 4 November 2026: cursor.com
  • Cursor docs: Agent, rules, skills, hooks, MCP, search, security, Agent Review, Design Mode and CLI, date read 4 November 2026: cursor.com
  • Cursor, "Cursor is now a part of SpaceX", 14 August 2026, date read 4 November 2026: cursor.com
  • Cursor, security notice on the Claude Code extension, 29 October 2026, date read 4 November 2026: cursor.com
  • Figma MCP server, remote server installation, date read 4 November 2026: developers.figma.com

If you're choosing a stack and want someone who designs the product and builds it with these tools, see my development work and book a free 30 minute intro call. After it, you get a written scope and one fixed quote.

Tags: Claude Code, Cursor, AI tools, Design engineering, Development

Keep reading

Let’s look at your product.

A free 30 minute call. You leave with a clear next step, whether we work together or not.

Book a free intro call