Cursor CLI brings the Cursor editor's agentic coding experience to the terminal—same modes, rules, MCP integrations, and model catalog, keyboard-first in your shell. Beyond natural-language prompts, it exposes 25+ slash commands for switching between Plan and Ask modes, configuring sandbox and auto-run behavior, managing rules and custom commands, and controlling sessions without leaving the CLI.
This article is a complete slash-command reference from the official Slash commands documentation and the Using Agent in CLI guide. It is not a replacement for upstream docs. Use this as a map of the surface area and a workflow guide; verify behavior against cursor.com/docs before you ship team runbooks.
Cursor CLI's slash commands orbit three modes—the same model as the Cursor editor:
Mode
Access
Switch via
Agent
Full tool access for coding tasks
Default; no flag needed
Plan
Research + clarifying questions + reviewable plan before code
/plan, Shift+Tab, --plan, --mode=plan
Ask
Read-only exploration; no file edits
/ask, --mode=ask
Plan mode works best for complex features, multi-file changes, unclear requirements, and architectural decisions. Ask mode is for understanding code without risk of edits. Agent mode is the default for implementation work.
See Plan Mode documentation for the full workflow: agent asks questions → researches codebase → generates plan → you review/edit → build when ready.
How slash commands fit a typical workflow
mermaid
flowchart LR
A[Setup /rules /mcp /sandbox] --> B[Plan /plan]
B --> C[Build Agent mode]
C --> D[Explore /ask]
D --> E[Ship /compress /usage]
E --> F[Sessions /new-chat /resume]
First session in a repo
Step
Command
What it does
1
/rules
Create or edit .cursor/rules for project guidance
2
/commands
Create or edit custom slash commands
3
/mcp list
Browse, enable, and configure MCP servers
4
/sandbox
Configure sandbox mode and network access
5
/auto-run status
Confirm tool auto-run policy
The CLI automatically loads AGENTS.md and CLAUDE.md at the project root alongside .cursor/rules—same as the editor agent.
During a task
Command
Purpose
/plan
Switch to Plan mode before large changes
/ask
Read-only codebase exploration
/model [model]
Set or list available models
/max-mode [on|off]
Toggle max mode on supported models
/auto-run [on|off|status]
Toggle whether tools run without confirmation
/compress
Summarize conversation to free context
/usage
View Cursor streaks and usage statistics
Before debugging or support
Command
Purpose
/copy-request-id
Copy last request ID to clipboard
/copy-conversation-id
Copy conversation ID to clipboard
/about
Show environment and CLI setup details
/feedback <message>
Share feedback with the Cursor team
Between sessions
Command
Purpose
/new-chat
Start fresh chat session
/resume <chat>
Resume previous chat by folder name
/logout
Sign out from Cursor
/quit
Exit CLI
Mode and model commands
Command
Purpose
/plan
Switch to Plan mode—design approach before coding
/ask
Switch to Ask mode—read-only exploration
/model [model]
Set active model or list available models
/max-mode [on|off]
Toggle max mode on models that support it
/auto-run [on|off|status]
Toggle auto-run default; status shows current state
Auto-run controls whether the agent executes tools (shell, file edits) without asking first. Tighten with /auto-run off when working on sensitive code; check state with /auto-run status.
Max mode enables extended reasoning on models that support it—similar in spirit to Claude Code's /effort or Codex's deeper reasoning tiers, but Cursor-specific.
CLI flags (non-slash alternatives):
bash
cursor agent --mode=plan "Refactor the auth module"
cursor agent --mode=ask "How does caching work?"
cursor agent --plan "Design a migration strategy"
Security, sandbox, and permissions
Command
Purpose
/sandbox
Configure sandbox mode and network access settings
/auto-run [on|off|status]
Control tool execution approval policy
Use /sandbox to restrict what the agent can reach on your machine and network—especially important when running Cursor CLI on shared or production-adjacent environments.
MCP integration matches the editor: enable servers with /mcp enable, disable with /mcp disable, browse config with /mcp list. See What is MCP? for protocol fundamentals.
/compress vs /new-chat:/compress keeps you in the same session with a summarized history—use mid-task when context fills up. /new-chat wipes context entirely—use when switching to an unrelated task.
Resume by folder:/resume <chat> finds previous sessions by folder name, aligning with Cursor's project-scoped chat history.
Debugging and support commands
Command
Purpose
/copy-request-id
Copy last request ID (for support tickets)
/copy-conversation-id
Copy conversation ID to clipboard
/feedback <message>
Send feedback to the Cursor team
/about
Dump environment and CLI configuration
When filing bugs or asking in Cursor forums, /copy-request-id gives support a precise trace of the failing request.
Complete slash command reference table
Every command from the official Slash commands page as of June 11, 2026. (The upstream docs list /max-mode twice; it is one command.)
Command
Purpose
/plan
Switch to Plan mode—design approach before coding
/ask
Switch to Ask mode—read-only exploration
/model [model]
Set or list models
/auto-run [on|off|status]
Toggle auto-run or show status
/sandbox
Configure sandbox mode and network access
/max-mode [on|off]
Toggle max mode on supported models
/new-chat
Start a new chat session
/vim
Toggle Vim keys for input
/help [command]
Show help; optional command name
/feedback <message>
Share feedback with the team
/resume <chat>
Resume previous chat by folder name
/usage
View Cursor streaks and usage stats
/about
Show environment and CLI setup details
/copy-request-id
Copy last request ID to clipboard
/copy-conversation-id
Copy conversation ID to clipboard
/logout
Sign out from Cursor
/quit
Exit the CLI
/setup-terminal
Auto-configure terminal keybindings
/mcp list
Browse, enable, and configure MCP servers
/mcp enable <name>
Enable an MCP server
/mcp disable <name>
Disable an MCP server
/rules
Create or edit agent rules
/commands
Create or edit custom slash commands
/compress
Summarize conversation to free context
Total: 24 unique built-in slash commands (25 entries in upstream table with duplicate /max-mode).
Five-way comparison: Cursor CLI vs Claude Code vs Codex vs Gemini CLI vs OpenCode
Cursor's differentiator:Plan + Ask + Agent mode triad with editor-parity .cursor/rules—a smaller built-in table than Claude Code, but tight integration with the Cursor ecosystem.
Commands worth memorizing first
bash
/plan # design before coding
/ask # read-only exploration
/model # pick model
/auto-run status # confirm tool policy
/compress # free context mid-session
/rules # edit project rules
/mcp list # check MCP servers
/new-chat # fresh session
/resume my-project # continue prior chat
/copy-request-id # for support tickets
Typical workflow
bash
# 1. Plan a large change
/plan
Refactor auth to use JWT with refresh tokens
# 2. Review plan, then switch to Agent (default) to implement# 3. Mid-session context getting full?
/compress
# 4. Explore without edits
/ask
How does the current session middleware work?
Cursor CLI slash commands center on three modes (/plan, /ask, Agent), editor-parity configuration (/rules, /commands, /mcp), and session hygiene (/compress, /new-chat, /resume). The built-in surface is smaller than Claude Code's 90+ commands, but the Plan → review → build workflow is first-class—not bolted on.
Command names and behavior reflect Cursor documentation as of June 11, 2026. Re-check cursor.com/docs and the Cursor changelog before documenting internal runbooks.
