What Is Claude Desktop? Features, Cowork Mode, Claude Code, and How to Get Started

What Is Claude Desktop?

Claude Desktop is the official native desktop application distributed by Anthropic for macOS and Windows. It packages the Claude chat assistant, Cowork mode (an in-app agent that can read and write local files), Claude Code (agentic coding), MCP connectors, and parallel session management into a single program that launches without a browser.

Think of Claude Desktop as the “AI hub for your laptop.” It signs in with the same Anthropic account as Claude.ai, syncs conversation history across devices, and exposes capabilities that simply are not available in the browser — most notably the ability to operate on the files on your machine through a sandboxed agent. For knowledge workers, that means generating PowerPoint decks from local Excel workbooks, reorganizing folders by natural-language request, or kicking off long-running Claude Code tasks that keep running while you move between apps.

Following the April 2026 redesign, Claude Desktop gained a session sidebar, drag-and-drop workspace, integrated terminal, and “Routines” (scheduled and event-triggered automations). Keep in mind that the redesign also consolidated Claude Code into the desktop app, so what used to be separate downloads are now one product.

How to Pronounce Claude Desktop

The “Claude” in Claude Desktop is pronounced like the French name — a single syllable rhyming with “awed.” It is important not to confuse it with the English word “cloud.” Anthropic’s official guidance is to use the French-style pronunciation consistently across markets.

How Claude Desktop Works

Claude Desktop is an Electron-based native application whose primary job is to act as a polished, OS-integrated front end for Anthropic’s Messages API. On top of that, it hosts the Claude Agent SDK runtime, which provides a sandboxed execution environment for tools like the filesystem reader, bash shell, and Cowork mode.

Architecture

Core Components

Claude Desktop breaks down into four core components. First, the chat UI, which mirrors Claude.ai’s conversational experience. Second, Cowork mode, which lets Claude read, write, and execute files in folders you explicitly grant access to. Third, Claude Code, which is the agentic coding environment tailored for software engineering workflows. Fourth, the MCP (Model Context Protocol) client, which lets you plug in external services such as Slack, Gmail, and Google Drive via a standard protocol. It is important to understand that each of these components runs through the same sandboxed Agent SDK, which enforces permission boundaries at the OS level.

Model Selection

As of April 2026, you can select between Claude Opus 4.6, Claude Sonnet 4.6, and Claude Haiku 4.5 from the model picker. A pragmatic rule of thumb: use Opus for complex multi-step agent work, Sonnet for everyday tasks, and Haiku for high-volume or latency-sensitive jobs. You should also note that some features — like Extended Thinking and long-context processing — are only available on specific models.

Session and Context Management

The redesigned sidebar treats every conversation as a first-class session. You can filter by status (active, paused, completed), group by project, or sort by recency. Running multiple sessions in parallel is expected behavior — the new “side chat” keyboard shortcut (Command + ; on macOS) lets you branch a clarifying question without polluting the main thread’s context. This matters because agent workflows burn context quickly, and the ability to spin up an isolated side conversation preserves the careful priming you invested in the main session.

Routines and Scheduled Execution

Routines package a prompt, target repository, and connector configuration into a reusable unit that can run on a cron-like schedule, trigger from an API call, or fire in response to GitHub events like a new pull request. Routines execute on Claude Code’s web infrastructure rather than your local machine, so they keep running even when your laptop is asleep. Daily run caps scale with your plan tier. Keep in mind that Routines are in research preview for Pro, Max, Team, and Enterprise users with Claude Code on the web enabled — expect both the feature set and the limits to evolve.

Claude Desktop Usage and Examples

Installing Claude Desktop is straightforward. Download the installer from the official site, run it, and sign in. From there, you unlock the full feature set by granting folder permissions to Cowork mode and configuring MCP servers.

Installation Steps

1. Visit claude.ai/download and pick macOS or Windows.
2. Run the installer (.dmg on macOS, .exe on Windows).
3. Sign in with your Anthropic account; Pro, Max, Team, or Enterprise is recommended.
4. Grant Cowork mode access to a working folder during the first-run setup.

Example: Configuring an MCP Server

// Example ~/.claude/claude_desktop_config.json
{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/Users/you/Documents"]
    },
    "slack": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-slack"],
      "env": {"SLACK_BOT_TOKEN": "xoxb-..."}
    }
  }
}

This snippet wires Claude Desktop up to a local Documents folder and Slack. Once restarted, Claude can read files in that folder, search Slack channels, post messages, and combine the two into higher-level tasks like “summarize this week’s design-team discussions and save the result as a Markdown file in Documents/reports/.”

Example: Using Cowork Mode

Point Cowork at a folder containing your Q1 sales spreadsheet, then type, “Analyze Q1 sales and produce a PowerPoint with three charts.” Claude will open the workbook, perform the analysis, call the pptx skill, and save a polished deck next to the source data. Note that Cowork cannot reach files outside the folders you granted — that boundary is enforced by the OS sandbox, not by a prompt.

Example: Parallel Claude Code Sessions

A realistic workflow for an engineer debugging two incidents at once: open session A and ask Claude Code to “reproduce the authentication failure in the staging logs since 14:00, propose a root cause, and draft a fix.” Open session B in the sidebar and ask it to “generate unit tests for the payment-retry branch and run them.” Claude Code works the two tracks in parallel, each with its own scratchpad and tool calls. The integrated terminal pane lets you run pnpm test or make lint without ever switching to another app. Keep in mind that parallel sessions share API rate limits, so budget accordingly on the Pro tier.

Example: Invoking a Skill

Skills live in ~/.claude/skills/. A skill is a folder with a SKILL.md manifest that tells Claude when and how to use it. Claude Desktop picks up skills automatically; dropping a new skill folder into ~/.claude/skills/ is enough to make it available. When your request matches a skill’s description, Claude loads the full skill content and follows its instructions. Anthropic ships reference skills for PowerPoint, Excel, Word, and PDF work that you can use as-is or clone and adapt.

Advantages and Disadvantages of Claude Desktop

Advantages

The productivity win is obvious: no browser context switch, always-on keyboard shortcuts, and file operations that would otherwise require copy-paste round-trips. MCP unifies a dozen everyday tools behind one conversation, and parallel sessions let you keep long-running agent jobs humming while you continue to work. For power users, the terminal-plus-diff-viewer combo makes Claude Desktop a viable replacement for a secondary IDE window.

Another underrated advantage is the unified authentication surface. Once signed in, every MCP connector, skill invocation, Cowork folder, and Claude Code session inherits the same identity. Users don’t juggle API keys for each tool, and admins don’t have to audit ten different auth flows. Keep in mind that this also makes onboarding new connectors nearly frictionless — you add a connector once and Claude can reason across all of them in a single prompt. For teams that already use Claude in the browser, migrating to Desktop adds capability without forcing a relearn of how to talk to the model.

Disadvantages

Electron carries a memory cost — expect 300–600MB of resident memory when idle, more with active sessions. Serious Cowork or Claude Code use requires a paid plan; the free tier is limited to chat. The mobile experience lives in separate Claude for iOS / Android apps, so if you move between laptop and phone constantly you still have to manage two surfaces. Keep in mind that Claude Desktop also requires the internet — there is no local-model fallback.

Startup latency is another honest trade-off. The first launch after a system reboot can take five to ten seconds as the Electron shell initializes and reconnects to MCP servers. If a connector is flaky (for example, an OAuth-expired Google Workspace link), the corresponding tool silently returns errors until you reauthenticate, which can confuse new users who expect a clear failure mode. Finally, because Cowork can touch local files, organizations with strict data-residency rules need to reason carefully about which folders they grant — it is important to establish that policy before, not after, rolling out the app.

Claude Desktop vs Claude.ai (Web)

Claude Desktop and Claude.ai share authentication and conversation history but diverge sharply on capabilities.

The short version: Claude Desktop optimizes for depth of integration with your laptop; Claude.ai optimizes for ubiquity. You should pick Desktop when you want Claude to act on your machine, and Web when you want access from any device.

In practice, most power users settle on a hybrid pattern. They keep Claude Desktop running on their primary workstation for file-heavy and code-heavy work, and fall back to Claude.ai on whatever device they happen to be using away from the desk. Conversation history synchronizes between both surfaces through the Anthropic account, so starting a brainstorm in the browser on a tablet and finishing it in Cowork on a laptop is seamless. Note that some features — parallel Claude Code sessions, terminal access, and Routines — are only available in Desktop, so plan your workflow accordingly.

Common Misconceptions

Misconception 1: Claude Desktop Runs Offline

Claude Desktop depends on Anthropic’s cloud-hosted models for inference, which means it requires a live internet connection. There is no bundled local model. If you want fully offline LLM usage, consider open-weight alternatives such as Llama 4 or Mistral running through Ollama or LM Studio.

Misconception 2: Claude Code and Claude Desktop Are Separate Products

Since the April 2026 redesign, Claude Code ships inside Claude Desktop as a first-class panel. The standalone CLI still exists, but the GUI workflow is the recommended path for most users. Important to note: older tutorials may describe them as distinct applications — that is out of date.

Misconception 3: Anthropic Sees All Files on Your Disk

Cowork mode only has visibility into folders you explicitly grant. The sandbox is enforced at the OS level, not by the prompt. Files you open obviously travel to the API for inference, but nothing outside the granted folders is ever touched. Enterprise plans additionally offer contractual data-training opt-outs.

Misconception 4: Skills and MCP Do the Same Thing

Skills and MCP are complementary, not substitutable. A skill is a prompt-and-instruction bundle (a SKILL.md file plus resources) that conditions how Claude approaches a task. MCP is a transport protocol that gives Claude tools it can call. The two frequently appear together: a skill for “generate a quarterly report” will typically invoke MCP tools to pull the underlying data. Important to note: you can write skills without any MCP servers, and you can use MCP without writing skills — but combining them is where Claude Desktop becomes genuinely agentic.

Misconception 5: Running Claude Code Needs a Separate Install

Historically Claude Code was a standalone CLI, which led to duplicate installs. Since the unified Desktop release, Claude Code runs inside Claude Desktop as a first-class panel; you do not need a separate npm install unless you specifically want the command-line workflow. This is especially relevant for teams migrating from the older CLI-only flow — the bundled panel preserves all CLI behavior while adding GUI affordances like diff viewers and session management. Keep in mind that plugins installed via the CLI and plugins installed via Desktop share the same on-disk registry, so installations in either surface become available to the other.

Real-World Use Cases

Sales teams use Cowork mode to auto-generate customer proposal decks from a CRM export. Engineering teams spin up three or four Claude Code sessions simultaneously — one refactoring a legacy module, another writing tests, a third reviewing a pull request. Marketers pipe GA4 metrics through the MCP analytics connector to produce weekly summaries. Finance teams run Routines that refresh a budget dashboard every Monday morning before standup.

Legal and operations teams have also found heavy use for Claude Desktop. In contract review, a reviewer grants Cowork access to a folder of vendor agreements, then asks Claude to cross-reference each document against the company’s standard redlines and flag deviations. Customer support leads pair Cowork with the Zendesk MCP connector to draft weekly escalation summaries from ticket exports. Research analysts use parallel sessions to compare two reports side-by-side while a third session synthesizes the findings into a memo. The common thread across these use cases is that Claude Desktop turns multi-step research and drafting chores — the kind of work that used to consume full afternoons — into conversational workflows that a human supervises rather than performs.

Representative Workflows

Typical workflows include: daily inbox triage with the Gmail MCP, meeting-note capture and action-item extraction, codebase onboarding (“summarize the auth module and list every TODO”), vendor contract review against a template, and Excel-to-slides conversion for recurring reports. Important: keep the scope of each task explicit — agent mode happily fans out when given ambiguous goals.

Enterprise Deployment Patterns

In enterprise deployments, Claude Desktop is often rolled out with organization-wide skills pre-installed via a central registry. Legal teams package contract-review procedures as skills; finance teams encode budget-modeling steps; engineering teams ship coding standards as reusable skills that every developer inherits. OpenTelemetry support lets ops teams ship usage telemetry into existing observability stacks, and role-based access controls let admins differentiate what sales reps, engineers, and executives can reach. Keep in mind that you should establish a clear policy on which folders and connectors are pre-approved before rolling out Cowork broadly.

Combining Desktop with Claude API

Desktop-first adopters often graduate to direct Anthropic API usage for production workloads. The pattern: prototype and iterate inside Claude Desktop, then port the prompt and tool setup to the Messages API for automated or customer-facing integrations. Extended Thinking, Prompt Caching, and Tool Use all work the same across Desktop and the raw API, so the transfer from exploration to production is mostly a matter of plumbing rather than re-architecture.

Frequently Asked Questions (FAQ)

Q. Is Claude Desktop free?

A. The Free tier includes basic chat. Cowork mode, Claude Code, and Routines require Pro (USD 20/mo) or higher. Max, Team, and Enterprise add higher limits and admin controls.

Q. Does Claude Desktop support languages other than English?

A. Yes — Claude’s multilingual capability is first-class, and the UI is localized into a growing list of languages, including Japanese. MCP servers and skills can be written and executed in any language Claude understands.

Q. Is it secure for enterprise use?

A. Team and Enterprise plans add SSO, RBAC, audit logs, and data-training opt-outs. Many regulated organizations (finance, healthcare) have deployed Claude Desktop in production. You should confirm your organization’s posture with the published SOC 2 and ISO 27001 reports.

Q. How does Claude Desktop compare to the ChatGPT desktop app?

A. The ChatGPT desktop app is primarily a chat surface. Claude Desktop’s distinguishing feature is Cowork mode — real filesystem and tool access through the Agent SDK — plus MCP connectivity. If you want an AI that works with your files and tools, Claude Desktop is the more agent-centric choice.

Q. Can I disable Cowork mode if I don’t need it?

A. Yes. Cowork mode is off by default and must be explicitly enabled per folder. You can revoke folder access at any time through the settings panel, and Claude loses all ability to see or write to that folder immediately.

Q. Does Claude Desktop work with a corporate proxy or VPN?

A. In general, yes — Claude Desktop respects system-level proxy configuration. For complex setups (TLS-intercepting proxies, custom certificate authorities), you may need to configure certificates at the OS level. Enterprise deployments often pre-configure this via MDM.

Q. How much context does Claude Desktop provide to each session?

A. It depends on the underlying model. Claude Opus 4.6 and Sonnet 4.6 offer 200K-token context windows, with larger extensions available on certain enterprise plans. Prompt Caching substantially reduces the cost of reusing long contexts across turns.

Q. Which operating systems are supported?

A. Claude Desktop ships native installers for macOS (Apple Silicon and Intel) and Windows (x86_64 and ARM64). Linux support is not officially offered as of April 2026, though some users run the Electron bundle under Wine for experimentation — this is unsupported.

Q. Can multiple users share one installation?

A. Each OS user profile gets its own Claude Desktop configuration, MCP connector list, and conversation history. On shared machines you should create separate profiles to keep sessions isolated. Enterprise plans add SCIM provisioning so IT can manage user lifecycle centrally.

Conclusion

Claude Desktop is the anchor application for Anthropic’s agent ecosystem on the personal computer. It brings the same models that power Claude.ai into a native shell that can reach into your filesystem, your terminal, your MCP-connected tools, and your IDE. For individuals it removes the browser as a bottleneck; for teams it centralizes skill distribution and connector governance; for enterprises it provides the audit and access-control surface that regulated industries require.

• Claude Desktop is Anthropic’s official native app for macOS and Windows.
• Cowork mode provides sandboxed local file and tool access — the headline capability.
• Claude Code, MCP connectors, parallel sessions, and Routines are bundled in one product.
• Pro or higher unlocks the full agent feature set; Free is chat-only.
• Internet is required; there is no local model fallback.
• Key differentiator vs Claude.ai: depth of OS integration, not raw model capability.
• April 2026 redesign added sidebar, terminal, diff viewer, and Routines.
• Skills, MCP, and Agent SDK are the three primitives that extend the product to any workflow.

References