Skip to content

Cursor ecosystem learning map

Cursor is Anysphere's coding-agent family: a local editor, a terminal CLI, cloud sandbox agents, a git forge, security review, visual UI direction, and an SDK. The fastest first hour is still Tab + Agent in the editor. This page is a map. It does not walk you through clicks.

Who this is for

  • Front-end / full-stack engineers moving from autocomplete to “let the agent edit the repo”
  • People who already know VS Code and are installing Cursor for the first time
  • Tech leads choosing among editor / CLI / Cloud / Bugbot / Origin / Security / Design / SDK

Not this page: enterprise contracts, per-token price grids, or Anysphere internals.

Product landscape

Official docs treat these as product surfaces, not another Rules recap.

Cursor family
├── Cursor editor (VS Code fork) — the local default
│   ├── Tab          — multi-line complete, cross-file jumps, auto-import
│   ├── Inline Edit  — Cmd+K / Ctrl+K on the current selection
│   ├── Chat / Agent — Cmd+I / Ctrl+I; one agent, four constraints
│   │   ├── Agent    — edit files, run commands
│   │   ├── Ask      — read-only Q&A
│   │   ├── Plan     — editable plan, then implement
│   │   └── Debug    — hypotheses → logs → you reproduce → evidence
│   └── Design Mode  — Agents Window browser: click / draw / voice (Cmd+Shift+D)
├── Cursor CLI (`agent`) — interactive terminal, or `agent -p` headless / CI
├── Cloud Agents (formerly Background Agents)
│   └── Isolated VMs: clone, branch, test, open PRs
│       Start from: editor Cloud dropdown, cursor.com/agents, mobile / PWA,
│       Slack / GitHub `@cursor`, or prefix a CLI message with `&`
├── Bugbot — automated PR review (bugs / security / quality)
│   └── Autofix spawns a Cloud Agent. Official name is Autofix, not a product called “Fixer”
├── Origin — Cursor's git forge (early beta). Host / mirror / PR / browse
├── Security Agents — Security Reviewer (PR) + Vulnerability Scanner (cron)
├── PR Routing & Approval — assign reviewers; approve low-risk PRs
└── Cursor SDK — `@cursor/sdk` / `cursor-sdk` / Bridge. Same agent, in your process

Also on the official nav (one line here, official pages — no extra Tutorial):

SurfacePick it whenOfficial
Mobile / iOS / PWAStart or review Cloud Agents from a phoneMobile
JetBrainsStay in IntelliJ / PyCharm / WebStorm via ACPJetBrains
XcodeXcode 26.3+ built-in MCPXcode
PluginsBundle rules / skills / MCP / hooksPlugins
AutomationsCloud Agents on a schedule or event. Hosts Bugbot, Security Agents, PR RoutingAutomations

Closest cousins on this site: Cloud Agents ≈ Claude remote / Dispatch and Gemini Jules; Design Mode ≈ Claude Design (in-app overlay vs standalone canvas); CLI ≈ Claude Code / Codex in a terminal; Origin ≈ “hosted code”, not Rules.

Quick decision: which should I use?

What do you want?
├── Finish the line / next edit / this selection
│   ├── Next edit → Tab (do not start an Agent turn)
│   └── This selection / function → Inline Edit (Cmd+K / Ctrl+K)
├── Ask, edit, plan, or debug in the local repo
│   └── Editor Agent (Cmd+I; Cmd+. for the mode menu)
│       ├── Understand first, do not touch files → Ask
│       ├── Clear multi-file work / tests → Agent
│       ├── Large or fuzzy change → Plan (Shift+Tab in the input)
│       ├── Reproducible bug, unknown cause → Debug
│       └── Point at the running UI → Design Mode (Cmd+Shift+D)
├── You are in a terminal, or a script / CI needs headless
│   └── Cursor CLI (`agent`)
│       ├── Interactive → `agent`
│       ├── Read-only → `agent --mode=ask`
│       ├── Plan first → `agent --mode=plan` / `--plan`
│       └── Script / CI → `agent -p`; add `--force` to apply edits
├── You are away, need parallelism, or want an isolated VM + PR
│   └── Cloud Agents
│       ├── Editor / web / iOS / Android PWA
│       ├── Slack, GitHub, Linear: `@cursor`
│       └── In a CLI session, prefix the message with `&`
├── Host the code itself (not just review a GitHub PR)
│   └── Origin (Pro / Teams / Enterprise; claim a codebase name)
├── Security scan or PR security gate
│   └── Security Agents (`/review-security`; Automations)
├── Route reviewers / auto-approve low-risk PRs
│   └── PR Routing & Approval (`APPROVAL_POLICY.md`)
├── Call the same agent from your TypeScript / Python / other runtime
│   └── Cursor SDK (`@cursor/sdk` / `cursor-sdk` / Bridge)
└── The PR is already pushed (or you want a diff review first)
    └── Bugbot
        ├── Hosted auto-review / comment `cursor review`
        ├── Local `/review-bugbot` before you push
        └── Machine should also patch → Autofix (Cloud Agent), not Debug Mode
You care aboutPickNext
IDE with Tab + Agent + rulesCursor editorTutorial
Click / draw / voice on the running UIDesign ModeDesign Mode
Terminal or headless CI, same Cursor rules / MCPCursor CLICLI
Away / parallel / isolated VM + PRCloud AgentsCloud Agents
Machine reviews the PR firstBugbotCookbook · Bugbot
Host / mirror / browse the repoOriginOrigin
Security bugs on PRs or on a cronSecurity AgentsSecurity Agents
Assign reviewers / approve low-risk PRsPR RoutingPR Routing
Same agent inside your processSDKCursor SDK
PhoneiOS app / Android PWAMobile
JetBrains / XcodeACP / Xcode MCPJetBrains · Xcode
Package skills + MCPPluginsPlugins
Schedule or event-driven Cloud runsAutomationsAutomations
Terminal, fine-grained permissions, Anthropic stackClaude CodeClaude map
Stay inside stock VS Code + GitHubGitHub CopilotCopilot page

The full capability matrix, shortcuts, and config templates live on the cheatsheet. This overview does not repeat that table.

Concepts in one line

Full “what / why” lives in the glossary.

ConceptOne lineWhere
TabCompletion model; accept / reject trains the next suggestionEditor
ModesAgent / Ask / Plan / Debug — constraints, not four productsEditor / CLI
Design ModeVisual prompts in the Agents Window browserEditor
Rules / AGENTS.mdPersistent instructions in model contextEditor / CLI / Cloud
Skills / CommandsOn-demand workflows; / to runEditor / Cloud
MCP / Hooks / SubagentsExternal tools, lifecycle scripts, isolated childrenEditor / CLI / Cloud (different limits)
Cloud AgentsAgents on isolated VMs; formerly Background AgentsCloud
BugbotPR reviewer; Autofix then spawns a Cloud AgentPR / /review-bugbot
CLI agentOfficial terminal entry; -p for headlessTerminal / CI
OriginCursor git forge; binary originHosted code
Security AgentsPR reviewer + cron scanner on AutomationsCloud / /review-security
PR RoutingAssign reviewers; maybe auto-approveAutomations
SDKSame agent from @cursor/sdk / cursor-sdk / BridgeYour process

Feature quick reference

Editor

FeaturePurposeDoc
TabCheap completions, no agent turnTutorial · Tab
Inline EditCmd+K on a selectionsame
Agent / Ask / Plan / DebugFour constraints on one local agentTutorial · Modes
Design ModeClick / draw / voice in the Agents Window browserDesign Mode
Rules / AGENTS.mdPersist package manager and folder conventionsTutorial · Project context

CLI / Cloud / review / host / SDK

FeaturePurposeDoc
agentSame agent in a terminalCLI
agent -p / --forceScripts and CI; --force to write filessame
Cloud AgentsRemote VM work and PRsCloud Agents
BugbotReviews a PR diff; default check is neutralCookbook · Bugbot
AutofixBugbot starts a Cloud Agent on findingssame
OriginHost, mirror, browse, PROrigin
Security AgentsPR + cron securitySecurity Agents
PR RoutingRoute / approvePR Routing
SDKIn-process agentCursor SDK

Learning path

Stage 1 — First useful change in five minutes

StepWhatLink
1Install, sign in, open a folderTutorial · Install
2Cmd+I: explain this repoTutorial · First example
3One low-risk edit, review the diff, run existing checkssame
4Tell Tab / Inline Edit / Ask / Agent / Plan / Debug apartTutorial · Modes

Stage 2 — Teach the project

StepWhatLink
1Root AGENTS.mdTutorial · Project context
2Split .cursor/rules/*.mdc by globCookbook · Rules
3.cursorignore for secretsCheatsheet · Ignore

Stage 3 — Real workflows

StepWhatLink
1Bugs / Debug ModeCookbook · Bugs
2Plan → implement → verifyCookbook · Features
3Bugbot on PRsCookbook · Bugbot
4MCP / Hooks / parallel agentsCookbook
5Visual UI editsDesign Mode
6LookupCheatsheet
7DefinitionsGlossary

Stage 4 — Leave the local window

StepWhatLink
1Install agent, run one interactive turnCLI
2Dispatch one Cloud taskCloud Agents
3Cloud ≠ Bugbot ≠ DebugGlossary · Cloud · Bugbot
4Host code on Origin, or scan / route PRsOrigin · Security Agents · PR Routing
5Embed the agentSDK

Built for frontend engineers · Powered by VitePress