Claude Desktop

Claude Desktop is centered on three tabs: Chat, Cowork, and Code.

In Chat, you do what you expect: converse with Claude, upload documents and images, and get answers. The app supports document and image analysis out of the box. You can drag in a PDF, a Word file, a spreadsheet, or an image, and Claude will read and discuss the contents. This works for any supported file type on your system, without needing to upload to a separate web portal.

Cowork is the agentic layer. Available on paid plans, Cowork lets you hand off multi-step tasks to Claude. You give it a goal — “Organize these receipts into a spreadsheet,” “Summarize this folder of research papers and generate a presentation” — and Claude executes autonomously. It runs code in an isolated VM on your machine, reads and writes files within connected folders, and can coordinate parallel workstreams. Cowork does not require a terminal; it is designed for non-technical knowledge workers but is powerful enough for developers.

Code is the development environment, bringing Claude Code into the desktop app. Each conversation is a session with its own project folder, chat history, and code changes. You can run multiple sessions in parallel with Git isolation. Within a session, Claude can edit files, run terminal commands, preview your app in an embedded browser, review diffs, and monitor pull requests through CI. You can also use “computer use” to let Claude control your screen, open apps, and click buttons — useful for testing or automation.

All three tabs rely on cloud models. There is no local model option. Your usage counts against your Claude plan’s rate limits. The app syncs conversations, projects, memory, and preferences across web, mobile, and desktop when logged into the same account.

Claude Desktop runs on macOS (version 11 Big Sur or later) and Windows (version 10 or later). There is no Linux build. Windows builds are available for both x64 and ARM64. Mac builds run natively on Apple Silicon and Intel. System requirements are modest: a computer that can run the OS, an internet connection for cloud inference, and sufficient disk space for the app itself (roughly 300 MB).

The app is free to download and use with a free Claude account. Pricing follows Claude’s plan tiers:

• Free: Limited usage, includes basic chat and document uploads. Cowork and Code are not available on the free plan.
• Pro ($20/month): Increased usage limits, access to Cowork and Code, faster models.
• Max ($100/month): Higher limits again, priority access, and extended context.
• Team and Enterprise: Per-seat pricing with administrative controls, SSO, and compliance features.

There are no hardware demands beyond a capable machine and a good internet connection. Because models run in the cloud, you do not need a powerful GPU or large RAM. This makes Claude Desktop accessible to users with integrated graphics, older Intel Macs, or budget Windows laptops — as long as they can run the OS.

The app is proprietary, closed-source. You cannot modify it or run custom models. It uses Anthropic’s API exclusively.

MCP Support (Model Context Protocol): Claude Desktop was the first product to ship MCP. With MCP, you can connect local servers that expose tools and data sources. Claude can then read files from your filesystem, query a local database, interact with a browser (via the Chrome extension), or use connectors for Google Drive, Slack, Excel, PowerPoint, Word, and more. MCP servers are installable as desktop extensions from a curated directory in Settings. Each extension uses code signing and encrypted storage for API keys. This is the feature that makes Claude Desktop viable for local tool and data workflows.

Native Desktop App: The app lives in your dock or taskbar. It supports a global keyboard shortcut (default Option+Space on Mac) for instant access. You can open Claude without switching browser windows or losing context. The window stays focused, and you can pin it to stay on top.

Document and Image Analysis: Upload PDFs, Word documents, Excel files, images, and more. Claude can extract text, understand diagrams, and answer questions about the content. This works locally — the file is uploaded to Anthropic’s servers for processing, but the action initiates from your machine.

Claude Cowork: As described, an agentic mode that automates multi-step knowledge work. It runs in an isolated VM, respects folder permissions, and can write files. You control which folders are accessible.

Claude Code: A full development environment integrated into the desktop app. Supports parallel sessions, Git integration, terminal, file editor, diffs, PR monitoring, and app previews. You can also send tasks from your phone to Claude Code on the desktop using Remote Control.

Cross-Device Sync: Conversations and memory sync across desktop, web, and mobile apps. You can start a conversation on your phone and continue on desktop.

Local Tool and Data Workflows — The primary use case for MCP. A data analyst can connect Claude to a local SQLite database and ask it to run queries. A writer can connect it to Google Drive and ask it to find all documents mentioning a specific project and create a summary. A developer can have Claude read a local codebase, identify issues, and open a PR — all from within the desktop app.

Deep Desktop Work — Writers, researchers, and project managers who need a dedicated AI assistant. Keep Claude open in its own window, upload reference material, and iterate on drafts or analysis without tab juggling. The global shortcut means you can invoke Claude from any app to ask a quick question.

Software Development — Using the Code tab, developers can run Claude Code sessions for code review, bug fixing, refactoring, or building new features. The ability to run multiple sessions with Git isolation is useful for exploring different approaches. Computer use can automate UI testing.

Connecting Your Own Services — Power users can wire Claude into their existing tool stack through MCP connectors. For example, connect a CRM, a ticketing system, or a BI tool. Claude can act on data from these services, generate reports, or trigger workflows.

Who Should Use It: Anyone who needs a cloud-powered AI assistant that can access local files and apps, especially professionals in writing, analysis, coding, and project management. Teams that want a consistent AI interface across desktop and mobile.

Who Should Look Elsewhere: Users who need offline or local models (try Ollama, LM Studio, or Jan). Developers who want open-source or customizable agent runners. Linux-only shops. Users who want native image generation (Claude can read images but not create them). Heavy users who hit plan limits — check Pro or Max tiers carefully.

1. Download: Go to claude.com/download and select the build for your operating system (macOS, Windows, or Windows ARM64).
2. Install: Open the downloaded file. On Mac, drag the app to your Applications folder. On Windows, run the installer. Launch the app.
3. Sign In: Log in with your existing Claude account. You can use a free account to start. No credit card is required for the free tier.
4. First Steps: Open a chat and upload a document or image. Try the global shortcut (Option+Space) to access Claude from anywhere.
5. Explore MCP: Go to Settings > Extensions. Browse the directory and install a few connectors: for example, the filesystem extension to let Claude read and write files, or the Chrome extension to let it control your browser.
6. Try Cowork (Paid): If you have a Pro or higher plan, open the Cowork tab and give Claude a task like “Find all PDF invoices in my Downloads folder and organize them by date in a new folder.”
7. Documentation: Full guides are at support.claude.com. For MCP, see “Getting Started with Local MCP Servers on Claude Desktop.”

Claude Desktop vs. ChatGPT for Desktop: Both are native apps for cloud-based models. ChatGPT for Desktop offers voice, image generation (DALL-E), and a similar native window. Claude Desktop wins on MCP and the ability to connect to local tools and files. If you need an AI that can read your local spreadsheet or query your database, Claude is the better choice. If you need image creation or prefer OpenAI’s models, ChatGPT is a strong alternative.

Claude Desktop vs. LM Studio / Ollama: These are local model runners. They let you download and run open-weight models completely offline, with no data leaving your machine. Claude Desktop cannot do that. If privacy, offline access, or zero API cost is your priority, choose LM Studio or Ollama. If you need the advanced reasoning and safety of Claude’s cloud models plus MCP integration, Claude Desktop is the tool. They do not directly overlap — one is a cloud GUI with local tool access, the other is a local inference engine.

Claude Desktop vs. Jan / Msty: Jan and Msty are also local-first desktop apps with support for multiple model providers (including cloud APIs). Jan is open-source and focuses on privacy. Msty offers a polished experience with RAG and web search. Claude Desktop is more limited in model choice (only Anthropic) but provides deeper integration with local tools through MCP and a more mature agentic layer (Cowork). For users who need only chat with cloud models, Jan or Msty offer more flexibility. For users who need automation and local file access, Claude Desktop is stronger.

• MCP support

  Connect local MCP servers so Claude can read files and use external tools and data sources from your machine.

• Native desktop app

  A standalone window with a global keyboard shortcut for quick access outside the browser.

• Document and image analysis

  Upload files and images and chat about their contents.