Cursor vs Claude Code: What I Found After Running Both

Both are installed on this machine. My repository has a .cursor/rules/ directory and a CLAUDE.md in the same commit, and on any given afternoon I use whichever window is already in front of me. That is an unsatisfying answer to cursor vs claude code, and it is the one most comparison articles avoid, because "different shapes for different work" does not make a headline. It is still what I have found to be true. Below is what I could verify about both tools with a source and a date attached, what developers running both say in public, and an honest read on who should pick which. I build MacMD Viewer, a $19.99 one-time read-only markdown viewer for macOS, which matters only at the end of this piece: whichever tool you pick fills your Mac with .md files neither tool is built to read comfortably.

TL;DR: Cursor is an editor you sit inside. Claude Code is an agent that runs in your terminal, your IDE, a desktop app and the browser. They overlap heavily and the vendors have stopped treating them as substitutes: Anthropic's docs ship an "Install for Cursor" link for its own extension, and Claude Code's /init reads .cursor/rules/. Pick on working posture, not on a benchmark, and re-check the prices yourself, because both vendors moved theirs inside the last quarter.

Every price and feature here was read from the vendors' own pages on 26 July 2026. There are no benchmark scores, deliberately: I could not tie a single one to an exact version, date and methodology, and both model lineups changed inside the last four weeks.

Cursor vs Claude Code: What Is the Actual Difference?

Cursor is a code editor with an agent inside it. Claude Code is an agent with editors attached. Everything else follows from that.

Cursor is made by Anysphere, Inc., founded in 2022, and it is a fork of VS Code (Wikipedia) — verifiable from version strings rather than folklore, since a Cursor staff moderator confirmed on 3 April 2026 that Cursor 3.0.9 bundles VSCode 1.105.1 (Cursor forum). One caveat on the "it's just VS Code" line, from that same reply: Cursor now uses its own markdown renderer rather than the native VS Code one. The fork has diverged where it touches documents.

Claude Code describes itself like this: "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. Available in your terminal, IDE, desktop app, and browser" (Anthropic). The flagship is the CLI; the same engine also runs as a VS Code extension, a JetBrains plugin, a desktop app, a web version and mobile apps, and "each surface connects to the same underlying Claude Code engine, so your CLAUDE.md files, settings, and MCP servers work across all of them."

The difference people report feeling is posture. The clearest statement of the CLI-first case I have read comes from a developer who switched:

"Being CLI-first is just so much more powerful than IDE-first, because tons of work that isn't just coding happens there. I use the VSCode extension in maybe 10% of my sessions when I want targeted edits." — physicles, Hacker News, 21 March 2026

And the opposite, from someone who prefers Cursor for exactly the reason the terminal wins for others:

"I much prefer Cursor to Claude Code. Claude Code hides too much of what it's doing from me (can't monitor the thinking output, can't see output from in-progress commands being run)." — x3n0ph3n3, Hacker News, 17 June 2026

Neither is wrong. They are two people describing what they want to see while a machine edits their code.

Is "Which One Should I Pick" Even the Right Question?

Often it is not, and the strongest evidence comes from the vendors rather than from anyone's opinion.

Anthropic ships a Claude Code extension that installs into Cursor, as a cursor: protocol URL sitting right beside the vscode: one on its own overview page (Claude Code overview). Traffic runs the other way too: Claude Code's /init "reads Cursor rules, in .cursor/rules/ or .cursorrules" and folds them into the CLAUDE.md it generates (memory docs). An importer for a competitor's config format and a one-click install into a competitor's editor is not how you treat a substitute.

The second reason to distrust a hard verdict is that this landscape does not hold still. Cursor's annualised revenue went from $100M in January 2025 to $3B in May 2026, and its valuation from $9.9B in June 2025 to $29.3B in November 2025 (via Wikipedia, citing TechCrunch and Bloomberg). Then Reuters reported on 16 June 2026 that SpaceX had agreed to buy Anysphere for $60 billion under a SpaceXAI subsidiary, completion expected in Q3 2026 — reported and pending, not closed, corroborated by the Hacker News thread from that day and by Cursor's own docs, which describe Grok 4.5 as "jointly trained by Cursor and SpaceXAI" (Cursor models). A valuation that roughly tripled across those five months, then a reported deal at double that figure seven months later, is a fair proxy for how fast this ground moves, and against it, 84% of developers now use or plan to use AI tools, up from 76% the year before (Stack Overflow 2025). A comparison of working styles survives that pace; a ranking does not.

How Do Cursor and Claude Code Compare, Feature by Feature?

Here is the matrix I actually use when someone asks. Prices read from the vendors' pages on 26 July 2026.

Dimension	Cursor	Claude Code
Core shape	Fork of VS Code, an IDE you open (3.0.9 bundles VSCode 1.105.1)	Agentic tool across five documented surfaces; the terminal CLI is the flagship
Where it runs	Desktop on macOS, Windows, Linux; iOS app in public beta since 29 June 2026; Slack; cloud agents	Terminal, VS Code, Cursor, JetBrains, desktop app, web, iOS and Android, Slack, GitHub Actions, GitLab CI
Editing paradigm	You sit in the editor, the agent edits around you; native Tab autocomplete	The agent proposes and you approve; inline diffs come via the IDE extensions
Own models	Composer 2.5, Composer 1, Grok 4.5	None; runs Anthropic's Claude models
Third-party models	Anthropic, OpenAI, Google, Z.ai, Moonshot	Claude only on subscription; Bedrock, Google Cloud and Microsoft Foundry routing for enterprise
Model routing	Cursor Router since 22 July 2026, with "Intelligence", "Balance" and "Cost" modes	Manual /model and /effort levels
Instruction file	.cursor/rules/*.mdc (plain .md is ignored) plus AGENTS.md	CLAUDE.md across four scopes plus .claude/rules/*.md; does not read AGENTS.md
Reads the other's config	—	/init reads .cursor/rules/ and .cursorrules
Plan artifacts	Plan mode; plans saved to your home directory, "Save to workspace" to commit them	Plan mode; Ctrl+G opens the plan in your editor
Extensibility	MCPs, skills and hooks; VS Code-style extensions; team marketplace	MCP, skills, hooks, subagents, agent teams, Agent SDK, plugins
Code review	Bugbot, usage-based billing	GitHub Code Review, GitHub Actions, GitLab CI
Entry paid tier	$20/month (Pro)	$17/month billed annually, $20 monthly (Pro)
Top individual tier	$200/month (Ultra)	Max, "From $100"/month, 5× or 20× Pro usage
Team seat	$40 (Standard) or $120 (Premium) per user per month	$20–25 (Standard) or $100–125 (Premium) per seat per month

Sources for the pricing rows: Cursor pricing docs and claude.com/pricing, both read 26 July 2026. Feature rows come from cursor.com/changelog, Cursor rules docs, Claude Code overview and permission modes.

One row deserves a correction rather than a cell, because it is the most repeated stale claim in this category: Cursor has MCP, skills and hooks, listed on the $20 tier (cursor.com/pricing). Comparison articles written in 2025 treat extensibility as a Claude Code exclusive, and it has not been true for a while.

Both tools also ship a plan mode, and the implementations differ where it counts. Cursor saves plans to your home directory by default, with a "Save to workspace" button to move one into the repository (Cursor planning docs). Claude Code gives you Ctrl+G at the approval prompt to open the proposed plan in your editor. If that gate is where you spend your attention, how Claude Code plan mode works covers the mechanics.

What Do Cursor and Claude Code Actually Cost?

Both start around $20 a month and both go far past it, and they meter so differently that a side-by-side monthly total would be fiction.

Cursor, from its pricing docs on 26 July 2026: free Hobby, Pro at $20/month, Pro Plus at $60, Ultra at $200, Teams at $40/user/month with Premium at $120, and custom Enterprise pricing. Its marketing page and its pricing docs currently give the same tiers different names, which tells you how recently they were rearranged.

The concept most articles skip: Cursor splits models into "Cursor Models" and "Other Models". Composer and Grok draw on a pool the pricing page calls "generous" and never quantifies. Third-party frontier models draw on a fixed dollar allowance — $20 on Pro, $70 on Pro Plus, $400 on Ultra — after which "on-demand usage allows you to continue using models after your included amount is consumed, billed in arrears" (cursor.com/pricing). Two developers on the same plan, one running Claude models and one running Composer, get very different bills.

Claude Code, from claude.com/pricing on the same date: Pro at $17/month billed annually or $20 monthly, which "Includes Claude Code"; Max at "From $100" per month, where you "Choose 5x or 20x more usage than Pro"; Team at $20/seat annually or $25 monthly for Standard and $100/$125 for Premium; Enterprise at $20/seat plus usage at API rates. The page shows only "From $100" for Max, so I will not quote a 20× figure I could not read there, however often it gets repeated elsewhere.

Claude Code meters time windows rather than dollars: "each member's Claude Code usage draws from a per-seat allowance that resets on a rolling five-hour window and a weekly window", shared with Claude chat and Cowork (Anthropic, costs). That page also carries the most useful number in this comparison, because it is a vendor publishing its own economics:

"Across enterprise deployments, the average cost is around $13 per developer per active day and $150-250 per developer per month, with costs remaining below $30 per active day for 90% of users."

Cursor publishes no equivalent, which is why there is no "typical monthly cost" table here. Half of it would be real and half would be my invention, and you could not tell which was which.

Re-read both pages yourself rather than trusting any article, including this one. On 21 April 2026 a 683-point Hacker News thread asked whether Claude Code was being removed from the Pro plan, on the evidence of people watching the page change under them: "5 minutes ago, I was seeing the old version of the page in which Claude Pro included Claude Code. I refreshed and now see that it does not" (tmp10423288442, comment). Today it reads "Includes Claude Code" on Pro again. Prices here have a shelf life measured in weeks.

What Do Developers Who Use Both Actually Say?

The public reports are genuinely split, and the split is not tribal. It tracks what people do all day.

For Cursor. Speed comes up repeatedly, including from people whose employers moved away from it: "My employer recently switched us all from Cursor to Claude Code … Claude Code is painfully slow compared to Cursor for my workload" (mnehring, 17 June 2026). So does the economics of Cursor's own models: "the composer model is basically unlimited and free also being super fast" (rustystump, 16 June 2026). And large repositories: "I use both a lot and find Cursor to be superior for the very large codebases I work in" (block_dagger, 2 September 2025). Put to a forced choice, some people who run both keep the editor: "At work I have both cursor and claude code and while I use both, cursor is by far the most pleasant to use. If I had to give one up, I'd let claude go" (jjav, 23 April 2026).

For Claude Code. The recurring theme is orchestration rather than editing. One developer who switched at the end of 2025 writes that "compared to Cursor this opened a different level of automation", describing swarms of agents running inside subscription limits (sermakarevich, 5 June 2026). Others report a category change rather than an increment — "a product that defined an entirely new category of product" (epistasis, 30 May 2026).

The cost story is the most common migration reason I found, and it comes from heavy users. One developer describes spending "around $2k a week with premium models" in Cursor before switching to Claude Code Max and being "equally as prolific and paying 1/10th the price" (dirtbag__dad, 3 April 2026). The fairest version concedes a point in each direction: "yes, the latter is noticeably slower with the same model at the same settings. However, it's hard to justify Cursor's cost. My bill was $1,500/mo at one point, which is what encouraged me to give CC a try" (enraged_camel, 24 April 2026).

And the dissent, which belongs here because a comparison that only quotes satisfied users is marketing. Agentic coding fails outright for some people on both tools: "I've been aggressively trying to use both Cursor and Claude Code … All of them resulted in sub-par results, essentially unusable" (exfalso, 3 August 2025). Stability complaints against Claude Code are recent: "I'd rather have my software not crash out on me every 5 minutes" (shimman, 20 July 2026).

Hold all of that next to one survey finding: developers distrust AI output more than they trust it. In the 2025 Stack Overflow survey, 46% actively distrust its accuracy against 33% who trust it, and 3.1% highly trust it. Whichever tool you run, the reading you do afterwards matters more than the choice between them.

Which One Should You Pick?

Three honest buckets, each tied to a reason someone actually gave rather than to a scorecard.

Pick Cursor if you still hand-edit a lot of code. Tab autocomplete is native and good, the agent works beside you rather than in front of you, and you can watch what it is doing without switching windows. If your budget pressure is per-request, Composer running on Cursor's own pool rather than the third-party dollar allowance is a real economic argument.

Pick Claude Code if the work surrounding your coding lives in a terminal. Log triage, git surgery, scripted one-offs, CI, piping output into a prompt. Anthropic's own example is tail -200 app.log | claude -p "Slack me if you see any anomalies", and that composability is not something an IDE-shaped tool reaches for. The same applies if you want several agents at once, or scheduled runs on Anthropic's infrastructure rather than on your laptop.

Pick either if you are new to this. The tie-breaker for a solo developer starting today is which interface you will actually open, not which is theoretically stronger. Both cost about $20 to start, both read a markdown instruction file at the repository root, and both are reversible in an afternoon.

One adoption data point exists and it needs its caveat attached. Among Stack Overflow's 2025 respondents who use or build AI agents — 8,323 people, 17% of the survey — 40.8% named Claude Code, behind ChatGPT at 81.7%, GitHub Copilot at 67.9% and Google Gemini at 47.4%. That is not 40.8% of all developers, and Cursor was not broken out in that question, so it says nothing about relative share.

How I Use Both

This is my workflow, not a recommendation and not a community standard. I went looking for evidence that developers converge on a particular split between these two and could not find it stated cleanly anywhere, so treat it as one person's arrangement.

I keep Claude Code in the terminal for anything spanning more than one file or needing to run commands, and Cursor open for the passes where I am typing in the file myself. The reason is unglamorous: my build, my git operations and my scripts already live in the terminal, so an agent there needs no bridge to reach them. For a small fix in a file I already have open, launching an agent is more ceremony than the change deserves.

What surprised me is how little friction running both costs. /init reading .cursor/rules/ meant my instruction files did not have to be written twice, and the Claude Code extension installs into Cursor when I want the agent inside the editor. Your split will depend on what your day is made of.

Where Does All That Markdown End Up?

On disk, in your repository, in volumes neither tool existed to help you read.

Cursor writes project rules to .cursor/rules/ as .mdc files, markdown with YAML frontmatter, and plain .md files there are ignored entirely (Cursor docs). It reads AGENTS.md from your project root, and saves plans to your home directory by default.

Claude Code writes CLAUDE.md across four scopes that concatenate rather than override, .claude/rules/*.md discovered recursively, and auto memory at ~/.claude/projects/<project>/memory/ with MEMORY.md as an index loaded every session. Anthropic's description of those memory files: "plain markdown you can edit or delete at any time" (memory docs).

The reading problem is shaped differently on each side. In Cursor, the Preview | Source toggle does not appear for *.plan.md files, files inside git worktrees, or files under .cursor/, .claude/ and .codex/ — precisely where agent output lands (Cursor forum, staff reply, 18 July 2026). In Claude Code the flagship surface is a terminal, so a plan arrives as scroll-back unless you press Ctrl+G to send it to an editor.

That is the gap MacMD Viewer fills, and it is the only claim I will make for it: a reading window, not a competitor to either tool. It renders any .md on disk regardless of path, reloads when an agent rewrites the file underneath it, draws Mermaid fences with no extension to install, and has no editing surface at all, so nothing in the reading window can write back into a file an agent is about to execute. QuickLook covers the rest — select a file in Finder, press Space, read it rendered.

Cursor users: the Cursor markdown viewer setup goes into the preview-toggle behaviour in detail. Claude Code users: reading plans before you approve them covers the approval gate.

Frequently Asked Questions

What is the difference between Cursor and Claude Code?

Cursor is a code editor with an agent inside it, forked from VS Code, and you work by sitting in the editor while the agent edits around you. Claude Code is an agentic tool that runs in a terminal, a VS Code or JetBrains extension, a desktop app, the web and mobile, and you work by describing a task and reviewing what it proposes. The overlap in capability is large; the difference in posture is what people actually feel.

Can you use Claude Code inside Cursor?

Yes. Anthropic's own overview page lists an "Install for Cursor" link for the Claude Code extension, as a cursor: protocol URL alongside the VS Code one. Claude Code's /init command also reads Cursor rules from .cursor/rules/ or .cursorrules when it generates a CLAUDE.md. The two products are not mutually exclusive at the install level.

Which is cheaper, Cursor or Claude Code?

They meter differently, so there is no honest single answer. Cursor gives each plan a dollar allowance for third-party frontier models ($20, $70 or $400 on Pro, Pro Plus and Ultra) and a separate "generous" pool for its own Composer and Grok models. Claude Code meters a rolling five-hour window plus a weekly window, shared with Claude chat. Anthropic publishes an average of "around $13 per developer per active day" for enterprise deployments; Cursor publishes no equivalent figure, so any head-to-head monthly table is half invented.

Does Claude Code read AGENTS.md or .cursor/rules?

It reads .cursor/rules, but only through /init, which folds Cursor rules into the CLAUDE.md it generates. It does not read AGENTS.md during a session. The documented workarounds are to put @AGENTS.md at the top of your CLAUDE.md or to symlink the two files. Cursor goes the other way: it reads AGENTS.md as plain markdown, and its own project rules must use the .mdc extension, since plain .md files in .cursor/rules/ are ignored.

Is Cursor still just VS Code with AI bolted on?

It is a VS Code fork, verifiably: Cursor 3.0.9 bundles VSCode 1.105.1 per a Cursor staff moderator in April 2026. But the fork has diverged where it matters for reading documents. The same staff reply states that Cursor uses its own markdown renderer rather than the native VS Code one, so VS Code advice about the preview pane does not reliably transfer.

Did SpaceX buy Cursor?

As of 26 July 2026 the deal is reported and pending, not closed. Reuters reported on 16 June 2026 that SpaceX had agreed to acquire Anysphere, Cursor's parent, for $60 billion, placing it under a SpaceXAI subsidiary, with completion expected in Q3 2026. Cursor's own model documentation already describes Grok 4.5 as "jointly trained by Cursor and SpaceXAI".

Which one is better for large codebases?

There is no verified answer, and the developer reports point both ways. One Hacker News commenter who uses both says Cursor is superior for the very large codebases they work in; another says being CLI-first is more powerful because so much non-coding work happens in the terminal. Both tools index your repository and both run agents against it, so the deciding factor in the reports is workflow shape rather than repository size.

The Short Version

The versus framing is the weakest part of this question. One vendor ships an install link for the other's editor, one reads the other's config format, and both pricing pages changed inside the last quarter. Pick the shape that matches where your work already happens: the editor if you still type in files, the terminal if your day is made of commands.

What does not change with the choice is the pile of markdown at the end of it, and how little attention it gets when 46% of developers do not trust what the agent wrote. Reading it is the cheap part. MacMD Viewer is $19.99 once and does that job and nothing else; the macOS markdown viewer roundup covers the alternatives if it is not the right fit.