Claude Code vs Cursor: slash commands and context syntax compared
Claude Code and Cursor's agent CLI both expose a slash-command layer on top of a chat prompt, and the two vocabularies overlap: /clear, /model, /plan, /goal, /fork, /rewind, /rename, /resume and /logout exist in both with the same meaning. The table below pairs each task with the literal command in each tool, taken from the two reference pages on this site and the vendors' command references; a dash means the tool has no documented equivalent.
| Task | Claude Code | Cursor |
|---|---|---|
| Start a fresh conversation | /clear [name] | /clear |
| Shrink the context window | /compact [instructions] | /summarize |
| Switch model | /model [model] | /model [filter] |
| Enter plan mode | /plan [description] | /plan [prompt] |
| Set a long-lived goal | /goal [condition|clear] | /goal [objective] |
| Fork the session | /fork [prompt] | /fork |
| Rewind to an earlier point | /rewind | /rewind |
| Rename the session | /rename [name] | /rename <name> |
| Resume a previous session | /resume [session] | /resume |
| Run a shell command from the prompt | ! | /shell [command] |
| Attach a file as context | @ | @auth.ts |
| Ask a side question | /btw [question] | — |
| Read-only question mode | — | /ask |
| Attach the branch diff | — | @Branch |
| Manage MCP servers | /mcp [reconnect (<server>|all)|enable|disable [<server>|all]] | /mcp [list|list-tools] [identifier] |
| Permissions / autonomy | /permissions | /run-everything [on|off|status] |
| Show cost or usage | /usage | — |
| Fix terminal newline keybinding | /terminal-setup | /setup-terminal |
| Sign out | /logout | /logout |
The differences are mostly about surface area. Claude Code's command list is several times longer because it bundles skills and workflows as slash commands (/code-review, /deep-research, /verify, /simplify) alongside built-ins such as /security-review, and exposes infrastructure switches (/hooks, /permissions, /remote-control). Cursor keeps the CLI list short. Cursor's @-mention family is richer at the prompt level, with @Branch, @Commit, @Terminals, @Chats and @Browser, where Claude Code's prompt layer has @ for file paths and ! for shell.
Three naming traps are worth knowing when switching. Context compaction is /compact in Claude Code and /summarize in Cursor; shell access is a leading ! in Claude Code but the /shell command in Cursor; the terminal newline fix is /terminal-setup in Claude Code and /setup-terminal in Cursor. Most of the rest either matches by name or has no counterpart, so the cost of moving between the two is small.
Full references: Claude Code (157) · Cursor (72) · all comparisons
Terminal agent versus editor agent
Claude Code runs in the terminal and treats the repository as its workspace; Cursor is an editor with the agent in a side panel and inline, plus a terminal agent that takes the slash commands listed here. The table pairs the actions that exist in both — starting, resuming and forking a session, switching model, planning, adding files as context — but the interaction model differs: Claude Code is a conversation in the terminal, while in the Cursor editor Ctrl + Enter (Cmd + Return on a Mac) accepts the changes the agent suggests and Tab accepts an inline completion.
Where the rows do not line up
Cursor's Tab completion and Cmd + K inline edit (Ctrl + K on Windows and Linux) are editor features with no Claude Code slash command to pair them with, so they are left to the Cursor shortcuts page. Inside the table four rows have a dash: /btw and /usage have no Cursor CLI command, and /ask and @Branch have no Claude Code one. Those rows are marked rather than stretched.