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 processAlso on the official nav (one line here, official pages — no extra Tutorial):
| Surface | Pick it when | Official |
|---|---|---|
| Mobile / iOS / PWA | Start or review Cloud Agents from a phone | Mobile |
| JetBrains | Stay in IntelliJ / PyCharm / WebStorm via ACP | JetBrains |
| Xcode | Xcode 26.3+ built-in MCP | Xcode |
| Plugins | Bundle rules / skills / MCP / hooks | Plugins |
| Automations | Cloud Agents on a schedule or event. Hosts Bugbot, Security Agents, PR Routing | Automations |
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 about | Pick | Next |
|---|---|---|
| IDE with Tab + Agent + rules | Cursor editor | Tutorial |
| Click / draw / voice on the running UI | Design Mode | Design Mode |
| Terminal or headless CI, same Cursor rules / MCP | Cursor CLI | CLI |
| Away / parallel / isolated VM + PR | Cloud Agents | Cloud Agents |
| Machine reviews the PR first | Bugbot | Cookbook · Bugbot |
| Host / mirror / browse the repo | Origin | Origin |
| Security bugs on PRs or on a cron | Security Agents | Security Agents |
| Assign reviewers / approve low-risk PRs | PR Routing | PR Routing |
| Same agent inside your process | SDK | Cursor SDK |
| Phone | iOS app / Android PWA | Mobile |
| JetBrains / Xcode | ACP / Xcode MCP | JetBrains · Xcode |
| Package skills + MCP | Plugins | Plugins |
| Schedule or event-driven Cloud runs | Automations | Automations |
| Terminal, fine-grained permissions, Anthropic stack | Claude Code | Claude map |
| Stay inside stock VS Code + GitHub | GitHub Copilot | Copilot 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.
| Concept | One line | Where |
|---|---|---|
| Tab | Completion model; accept / reject trains the next suggestion | Editor |
| Modes | Agent / Ask / Plan / Debug — constraints, not four products | Editor / CLI |
| Design Mode | Visual prompts in the Agents Window browser | Editor |
| Rules / AGENTS.md | Persistent instructions in model context | Editor / CLI / Cloud |
| Skills / Commands | On-demand workflows; / to run | Editor / Cloud |
| MCP / Hooks / Subagents | External tools, lifecycle scripts, isolated children | Editor / CLI / Cloud (different limits) |
| Cloud Agents | Agents on isolated VMs; formerly Background Agents | Cloud |
| Bugbot | PR reviewer; Autofix then spawns a Cloud Agent | PR / /review-bugbot |
CLI agent | Official terminal entry; -p for headless | Terminal / CI |
| Origin | Cursor git forge; binary origin | Hosted code |
| Security Agents | PR reviewer + cron scanner on Automations | Cloud / /review-security |
| PR Routing | Assign reviewers; maybe auto-approve | Automations |
| SDK | Same agent from @cursor/sdk / cursor-sdk / Bridge | Your process |
Feature quick reference
Editor
| Feature | Purpose | Doc |
|---|---|---|
| Tab | Cheap completions, no agent turn | Tutorial · Tab |
| Inline Edit | Cmd+K on a selection | same |
| Agent / Ask / Plan / Debug | Four constraints on one local agent | Tutorial · Modes |
| Design Mode | Click / draw / voice in the Agents Window browser | Design Mode |
Rules / AGENTS.md | Persist package manager and folder conventions | Tutorial · Project context |
CLI / Cloud / review / host / SDK
| Feature | Purpose | Doc |
|---|---|---|
agent | Same agent in a terminal | CLI |
agent -p / --force | Scripts and CI; --force to write files | same |
| Cloud Agents | Remote VM work and PRs | Cloud Agents |
| Bugbot | Reviews a PR diff; default check is neutral | Cookbook · Bugbot |
| Autofix | Bugbot starts a Cloud Agent on findings | same |
| Origin | Host, mirror, browse, PR | Origin |
| Security Agents | PR + cron security | Security Agents |
| PR Routing | Route / approve | PR Routing |
| SDK | In-process agent | Cursor SDK |
Learning path
Stage 1 — First useful change in five minutes
| Step | What | Link |
|---|---|---|
| 1 | Install, sign in, open a folder | Tutorial · Install |
| 2 | Cmd+I: explain this repo | Tutorial · First example |
| 3 | One low-risk edit, review the diff, run existing checks | same |
| 4 | Tell Tab / Inline Edit / Ask / Agent / Plan / Debug apart | Tutorial · Modes |
Stage 2 — Teach the project
| Step | What | Link |
|---|---|---|
| 1 | Root AGENTS.md | Tutorial · Project context |
| 2 | Split .cursor/rules/*.mdc by glob | Cookbook · Rules |
| 3 | .cursorignore for secrets | Cheatsheet · Ignore |
Stage 3 — Real workflows
| Step | What | Link |
|---|---|---|
| 1 | Bugs / Debug Mode | Cookbook · Bugs |
| 2 | Plan → implement → verify | Cookbook · Features |
| 3 | Bugbot on PRs | Cookbook · Bugbot |
| 4 | MCP / Hooks / parallel agents | Cookbook |
| 5 | Visual UI edits | Design Mode |
| 6 | Lookup | Cheatsheet |
| 7 | Definitions | Glossary |
Stage 4 — Leave the local window
| Step | What | Link |
|---|---|---|
| 1 | Install agent, run one interactive turn | CLI |
| 2 | Dispatch one Cloud task | Cloud Agents |
| 3 | Cloud ≠ Bugbot ≠ Debug | Glossary · Cloud · Bugbot |
| 4 | Host code on Origin, or scan / route PRs | Origin · Security Agents · PR Routing |
| 5 | Embed the agent | SDK |