Module 1 of 6
Meet the Claude Clients
"Claude" isn't one app — it's a family of surfaces built on the same models. This module explains what Claude.ai chat, Claude Desktop, and Claude Code each are for, and gives you concrete rules of thumb for picking the right one.
The three clients, at a glance
Every Claude client sends your messages to the same underlying models, but each one wraps that conversation in a different environment with different tools attached. The right choice depends on where your work already lives and what Claude needs to touch to help you.
| Client | What it is | What it's best for |
|---|---|---|
| Claude.ai (chat, browser) | A web app for chatting with Claude, uploading files/images, and using Projects, Artifacts, and connected apps. | Quick questions, drafting and editing documents, brainstorming, analyzing a document or spreadsheet you upload, and sharing a polished result (an Artifact) with someone else. |
| Claude Desktop (native app) | A native app version of the Claude.ai experience that also supports local MCP (Model Context Protocol) connectors, so Claude can reach tools and data on your machine or in your internal systems. | Everyday chat work that also needs access to local files or internal tools via MCP — e.g. querying an internal system, reading local documents, or using company-specific integrations that aren't available in the browser. |
| Claude Code (CLI / IDE agent) | A command-line and editor-integrated agent that reads and edits files in a real project directory, runs shell commands, uses git, and can work autonomously across many steps. | Working directly in a codebase or file tree: writing and refactoring code, running tests and builds, multi-step automation, and any task where Claude needs to actually do things on your filesystem, not just talk about them. |
What each one is actually for
Claude.ai chat
Think of Claude.ai as your default starting point for conversational work. You type or paste in what you need, Claude responds in the chat, and if the output is something substantial (a document, a slide outline, code you want to review as a whole), it can render as an Artifact you can view, edit, and share as a link.
It has no access to your local machine — everything happens in your browser tab, scoped to what you upload or paste, plus whatever connected apps/integrations you've enabled.
Claude Desktop
Claude Desktop looks and feels like Claude.ai, but as a native app it can be configured with MCP servers — connectors that let Claude read local files, query internal databases, or call company tools directly from the conversation.
Use it when you want the familiar chat experience, but the task needs Claude to reach something Claude.ai's browser sandbox can't: a local folder, an internal API, or another desktop tool.
Claude Code
Claude Code is an agentic coding tool that runs in your terminal or editor. It operates on a real project directory: it can read and search files, make edits, run builds and tests, use git, and chain many tool calls together to complete a task with minimal hand-holding.
It's built for software and file-system work — not just discussing code, but actually changing it, verifying the change, and iterating.
When to use which
Use Claude.ai chat when...
- You have a quick question or want to brainstorm ideas.
- You want to draft, edit, or polish a document, email, or plan.
- You need to analyze a file or image you can just upload.
- You want a shareable Artifact (a doc, slide outline, or small app) to hand to a colleague.
Use Claude Desktop when...
- You want the same chat experience as Claude.ai, but need Claude to reach local files or internal tools via MCP.
- Your team has set up connectors to internal systems (tickets, wikis, databases) that only work through desktop MCP integrations.
- You're doing recurring knowledge work that benefits from persistent local context, not one-off code changes.
Use Claude Code when...
- You're working inside an actual codebase or project folder.
- The task requires editing multiple files, running tests, or using git.
- You want an agent that plans and executes multi-step technical work with limited supervision.
- You need reproducible, auditable changes (diffs, commits) rather than a copy-pasted chat answer.
A common mistake
Pasting large chunks of code into Claude.ai chat to ask for a fix, then manually copying the answer back into your editor. If you're iterating on a real project, Claude Code will read the actual files, make the edit in place, and can verify the change by running your tests — saving the copy/paste round-trip and reducing the chance of losing context.
Quick decision rule
Ask yourself: "Does this task live in a project directory that needs files changed and commands run?"
- Yes → Claude Code.
- No, but I need Claude to reach local files or internal tools from a chat conversation → Claude Desktop.
- No, it's a conversation, document, or one-off analysis → Claude.ai chat.
Want to see the difference for yourself?
Head to the Try It playground and type the same request into simulated versions of all three clients — it's the fastest way to feel why each one behaves the way it does.