Claude Surfaces & Features – Developer Navigation Guide

Claude exists across multiple surfaces, each with distinct capabilities, scopes, and target audiences. This guide maps the landscape for developers who need to know where to work and why it matters.

1. The Three Primary Surfaces

Surface	Interface	Who it's for
Claude.ai Chat	Browser / web app	Everyone – quick questions, writing, analysis
Claude.ai Projects	Browser / web app	Organized, persistent, team-shared workspaces
Claude Code	Terminal / IDE extensions	Developers – coding, agentic tasks, automation

These are not modes of the same app. They are separate products with different capabilities and different persistent state.

2. Chat vs. Projects: What's the Difference?

Chat (claude.ai regular conversations)

A regular chat is stateless across sessions. Claude has no memory of your previous chats. Each conversation starts fresh. Good for:

One-off questions
Quick drafts or translations
Exploring ideas without wanting prior context to interfere

Projects (claude.ai → "New Project")

A Project is a persistent workspace with:

System instructions – A prompt that applies to every conversation in this project
Uploaded knowledge – PDFs, code files, docs that Claude can reference
Conversation history – All conversations inside a project share context over time
Scoped connectors – MCP connectors that are active only within this project

Think of a project as a "pre-configured Claude instance" for a recurring domain (e.g., a "BSH Compliance Research" project with our regulatory docs uploaded, or a "Saurus Docs" project with the repo context).

Team Projects

On Teams and Enterprise plans, projects can be shared across the team. Every team member gets the same system instructions and knowledge base — consistent Claude behavior without copy-pasting prompts.

When to use which

Scenario	Use
One-off draft, translation, brainstorm	Chat
Recurring domain (compliance, support, product research)	Project
Coding tasks, refactoring, CI automation	Claude Code
Team needs consistent AI persona/knowledge base	Shared Project

3. Claude Code: Routines vs. Schedule

Claude Code adds two distinct scheduling mechanisms. They sound similar but operate differently.

Routines (Hooks)

Hooks are shell commands wired to Claude Code events — they fire automatically when something happens in a Claude Code session:

// .claude/settings.json
{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Bash",
        "hooks": [{ "type": "command", "command": "npx eslint --fix $FILE" }]
      }
    ]
  }
}

Hooks are event-driven, local, and synchronous with Claude's actions. They are not scheduled by time — they fire when Claude does something.

Schedule (Scheduled Remote Agents)

/schedule creates cron-based remote agents that run on Anthropic's infrastructure on a time interval, independently of whether you have a Claude Code session open:

/schedule every day at 09:00 – check for new GitHub issues and summarize them in Slack

Runs in the cloud, not on your machine
Persists across restarts
No terminal session required
Managed via /schedule list / /schedule delete

Quick comparison

	Hooks (Routines)	Schedule
Trigger	Claude Code event	Time (cron)
Runs on	Your machine	Anthropic cloud
Session required	Yes	No
Use case	Lint on save, format on edit	Daily summaries, nightly checks

4. Connectors: Why They Appear in Three Places

You will find "connectors" in three distinct locations in the Claude.ai UI. This is intentional — each scope does something different.

4.1 Global Connectors (`Settings → Connectors`)

Available in every conversation and every project across your account. Use this for tools you always want — e.g., a search connector, a calendar connector.

4.2 Project-Scoped Connectors (Project → Settings → Connectors)

Available only inside a specific project. Useful when a connector is only relevant in one domain (e.g., a Jira connector active only in your "Sprint Planning" project, not in your "Customer Emails" project).

4.3 Plugins (within a conversation)

"Plugins" is Anthropic's branding for curated first-party or verified connectors that appear in the conversation toolbar. These are the same MCP-based connectors under the hood — they are just surfaced differently (rendered UI, one-click activation). Examples: Figma, Canva, Asana.

4.4 The MCP Connector Developer Guide

The MCP Connector Guide is for building your own connector — writing the server code, deploying it, and registering it as a custom connector in any of the scopes above.

Mental model

Plugins          →  curated first-party, rendered UI, in-chat toolbar
                        ↓ same MCP protocol underneath
Global Connectors  →  your registered MCP servers, available everywhere
Project Connectors →  same, but scoped to one project
Custom Connector   →  you built the MCP server (see the Dev Guide)

5. As a Developer: Claude Desktop vs. Claude Code

This is the most important question for daily workflow.

Claude Desktop (the native GUI app)

Claude Desktop is a native Mac/Windows application that wraps the claude.ai chat experience. It adds:

Local MCP server support via claude_desktop_config.json
Persistent conversations (same as claude.ai Projects)
Works offline for the UI (but model calls still need internet)

Target audience: Non-developers who want local MCP tools (Notion, Slack, Calendar) without touching a terminal.

Limitations for developers:

No codebase awareness — it does not know what's in your project
No git integration
No hooks, no routines, no /slash commands (beyond built-ins)
Cannot run tests, linters, or build commands
MCP config is per-machine, not per-project
No IDE integration

Claude Code (Terminal / IDE)

Claude Code is a developer CLI designed to operate inside your project's context. It understands:

Your entire codebase (reads files, navigates the tree)
Git state (current branch, diff, log)
Running processes (dev servers, test runners)
Your terminal environment

Developer-specific features:

Feature	Claude Code	Claude Desktop
Codebase reading	✅ Full tree	❌
Git awareness	✅	❌
Run shell commands	✅	❌
Hooks / event automation	✅	❌
`/schedule` remote agents	✅	❌
MCP tools (local)	✅ `claude mcp add`	✅ config file
MCP tools (remote/HTTP)	✅	Limited
IDE integration	✅ VS Code, JetBrains	❌
Project-scoped config	✅ `.claude/settings.json`	❌
Custom slash commands	✅ `SKILL.md`	❌
CLAUDE.md instructions	✅	❌

Skip Claude Desktop if you're a developer

Claude Desktop is not a developer tool. If your work involves code, a terminal, or a project directory, Claude Desktop cannot help you there. Use Claude Code in the terminal or via the VS Code / JetBrains extension instead.

When Claude Desktop is still useful for developers

Non-code domains in a GUI context: Research sessions, writing, document review — where you want a polished GUI without a terminal.
Local MCP testing (quick): Verifying a locally built MCP server works before deploying — faster to wire up in claude_desktop_config.json for a smoke test than full Claude Code MCP setup.
Non-technical colleagues: Handing off a configured Desktop setup to a PM or designer who needs local tools.

The recommended developer setup

Daily coding work      →  Claude Code in terminal / IDE extension
Project automation     →  Claude Code + Hooks + /schedule
Quick questions        →  claude.ai in browser (or claude.ai Projects for your domain)
Local MCP smoke test   →  Claude Desktop (then switch to Claude Code for real use)

6. MCP Scope Matrix

Here's the full picture of where MCP servers can be registered and what they can reach:

MCP Scope	Registered in	Available in	Per-project config
Claude Code – user scope	`claude mcp add -s user`	All Claude Code sessions	❌ (global)
Claude Code – project scope	`claude mcp add -s project`	This project only	✅ `.mcp.json`
Claude.ai – global connector	`Settings → Connectors`	All chats + projects	❌
Claude.ai – project connector	`Project → Settings → Connectors`	This project only	✅
Claude Desktop	`claude_desktop_config.json`	Desktop app only	❌ (machine-wide)

7. Quick Decision Tree

I want to...
│
├── Chat with Claude quickly, no setup
│   └── → claude.ai Chat
│
├── Give Claude persistent context for a recurring domain
│   └── → claude.ai Project (+ upload docs, set instructions)
│
├── Share a Claude workspace with my team
│   └── → claude.ai Shared Project (Teams/Enterprise plan)
│
├── Work on code, run commands, understand my repo
│   └── → Claude Code (terminal or IDE extension)
│
├── Automate a recurring task on a schedule
│   └── → Claude Code + /schedule
│
├── Auto-trigger actions when Claude writes code
│   └── → Claude Code + Hooks
│
├── Build a custom MCP connector
│   └── → See MCP Connector Dev Guide, deploy as custom connector
│
└── Just want a GUI, no terminal, with local MCP tools
    └── → Claude Desktop (note: not for active dev work)

All AI-generated content is a draft and requires human review before external use.

1. The Three Primary Surfaces​

2. Chat vs. Projects: What's the Difference?​

Chat (claude.ai regular conversations)​

Projects (claude.ai → "New Project")​

When to use which​

3. Claude Code: Routines vs. Schedule​

Routines (Hooks)​

Schedule (Scheduled Remote Agents)​

Quick comparison​

4. Connectors: Why They Appear in Three Places​

4.1 Global Connectors (Settings → Connectors)​

4.2 Project-Scoped Connectors (Project → Settings → Connectors)​

4.3 Plugins (within a conversation)​

4.4 The MCP Connector Developer Guide​

Mental model​

5. As a Developer: Claude Desktop vs. Claude Code​

Claude Desktop (the native GUI app)​

Claude Code (Terminal / IDE)​

When Claude Desktop is still useful for developers​

The recommended developer setup​

6. MCP Scope Matrix​

7. Quick Decision Tree​