Switching from Cursor to Claude Code: what changes

When you switch from Cursor to Claude Code, two things move: your rules go into CLAUDE.md (or .claude/rules/ for file-specific ones), and your MCP servers go into .mcp.json or claude mcp add-json. What changes is the way you work: you give an agent tasks instead of accepting autocomplete. And you don't have to leave Cursor at all.

This page is the move, file by file. Every config detail below is from Cursor's and Anthropic's own docs, read on 22 October 2026. If you're still deciding whether to move, Cursor vs Claude Code compares the two.

What moves when you switch from Cursor to Claude Code#

Move your rules#

Claude Code doesn't read .cursor/rules. It reads CLAUDE.md, and from version 2.1.277 it also reads AGENTS.md when your repository has no CLAUDE.md (Claude Code docs). Cursor supports AGENTS.md too, so if you already keep one, both tools can share it as it is.

Otherwise, move them in four steps:

1. Commit first, so the move is one clean diff.
2. In the repo, run claude, then ask: "Read every file in .cursor/rules. Put the always-apply rules in CLAUDE.md. Turn each rule with globs into a file in .claude/rules/ with a paths: list in its frontmatter. List any rule you weren't sure how to move."
3. Read the result. Anthropic's advice is to keep CLAUDE.md short: for each line, ask "Would removing this cause Claude to make mistakes?"
4. Start a new session and run /context to check the files loaded.

A path-scoped rule looks like this, from Anthropic's docs:

---
paths:
  - "src/api/**/*.ts"
---

# API Development Rules

- All API endpoints must include input validation

It loads only when Claude reads or edits a matching file, much like a Cursor rule with globs.

Move your MCP servers#

Good news: the two formats are close. Cursor keeps servers in .cursor/mcp.json (project) or ~/.cursor/mcp.json (global), under an mcpServers key (Cursor docs). Claude Code's project file, .mcp.json at the repo root, uses the same mcpServers key (Claude Code docs).

1. Project servers. Copy .cursor/mcp.json to .mcp.json. Local servers (command, args, env) carry over as they are. For each remote server with a url and no type, add "type": "http" (or "sse" or "ws", to match the server). Claude Code reads an entry with no type as a local server, so a url entry without one fails.
2. Global servers. Add each one with claude mcp add-json <name> '<json>' --scope user, pasting the object from inside mcpServers. Use a name made of letters, numbers, hyphens and underscores only, as Anthropic's docs ask. User scope makes it available in every project.
3. Sign in again. Servers that use OAuth need a fresh sign-in in Claude Code. Type /mcp, and check each server shows as connected.

If you're picking servers fresh, our list of the best MCP servers for Claude Code is a short place to start.

What you lose, and what you gain#

Be honest with yourself about the first list before you switch.

What you lose

• Tab. Cursor describes it as "Cursor's AI-powered autocomplete" that "suggests code as you type" (Cursor docs). Claude Code has nothing like it. If you write a lot of code by hand, you'll miss it.
• Edit a selection. In Cursor, Cmd/Ctrl+K on selected code now opens Agent with that code; Cursor's docs say it "used to open inline edit, which has been replaced by this workflow" (Cursor docs). The closest thing in Claude Code is selecting code with the VS Code extension and asking about it.
• One app for everything. Cursor is the editor and the agent. Claude Code is the agent; you bring the editor.

What you gain

• Your Claude plan covers it. As of 22 October 2026, Claude Pro ($20 a month) and Max include Claude Code. If you already pay Anthropic, there's no second subscription.
• Hooks, skills and subagents. Scripts that always run at set points (tests before Claude may stop, a linter after each edit), reusable workflows, and helpers with their own context.
• It runs anywhere a terminal does. Over SSH, in tmux, in CI with claude -p, on a server with no screen.
• Checkpoints. Rewind the conversation, the code, or both, with Esc twice or /rewind.

Keep Cursor, run Claude Code inside it#

Switching agents doesn't mean switching editors. Two ways to have both:

• In Cursor's terminal. Open the integrated terminal and run claude. It works in any terminal.
• With the extension. Anthropic's VS Code docs give an install link for Cursor as well as VS Code (Claude Code docs). You get a chat panel, your selection as context, and, in Manual mode, Claude's proposed edits as side-by-side diffs to accept or reject.

You can split the work: Cursor for reading and small edits by hand, Claude Code for the tasks. If the CLI is new to you, Claude Code for beginners takes you from the install to a first task and the five commands that matter.

Open in Cursor from Shuriken#

Shuriken runs Claude Code, Codex and other agents in an app, with every new chat on its own branch in its own git worktree. That raises a fair question: where is this chat's code, and how do I look at it in Cursor?

In the desktop app, every chat has Open in. Pick Cursor, and Shuriken opens that chat's worktree, or a file at its spot, in Cursor. It lists only the editors installed on that computer, VS Code, Zed and the JetBrains IDEs among them; the full list is in Open anywhere. You can also run Cursor's own CLI agent as a chat, with your Cursor sign-in. How the two compare as places to run agents is on Shuriken vs Cursor agents.

If you'd like Claude Code chats that open straight into Cursor, download Shuriken.