Everything Claude can do.
Most people use 20% of it.

Projects give Claude a persistent memory and a fixed context that carries across every conversation inside that project. You upload files, set a system prompt (custom instructions), and Claude remembers everything — your writing style, your company background, your preferences — without you having to re-explain it each time.

Create separate Projects for different areas of your work: one for client A, one for your newsletter, one for code review. Each has its own documents, instructions, and conversation history. Claude in each project feels like a specialist who knows your stuff.

Memory is separate — it's Claude remembering facts about you across all conversations, not just one project. Things like your job, your timezone, how you prefer to receive feedback. You can see and edit what Claude has remembered.

Extended thinking is a mode where Claude works through a problem step by step before giving you a final answer — like a scratchpad it thinks on before committing. You can see the reasoning process. It takes longer, but the answers are noticeably better for hard problems.

This is particularly powerful for: maths and logic, complex decisions with many trade-offs, writing where structure matters (arguments, strategy docs), debugging tricky code, and any analysis where you need to see the reasoning not just the conclusion.

Think of it as the difference between Claude answering immediately and Claude actually thinking it through. For anything consequential, use extended thinking.

When Claude creates something — a document, a piece of code, a diagram, a spreadsheet, a small web app — it can render it in a panel beside the conversation rather than dumping it into the chat. You can edit, preview, copy, and download it directly.

This makes Claude genuinely useful for producing real outputs rather than just text to copy and paste. Ask it to build you a simple calculator — it renders a working one you can interact with instantly. Ask for a formatted document — it appears styled, not as raw markdown.

Artifacts support: React components, HTML pages, SVG diagrams, Markdown documents, code files, and more. You can iterate on them — "make the button blue, add a third column, change the tone in paragraph 2" — and watch it update in place.

Claude can now generate images natively, without switching to a separate tool. The image generation is tightly integrated — Claude understands context from the conversation, so you can say "generate a logo for the brand we've been discussing" and it knows what you mean.

What sets Claude apart from standalone image tools: the conversation context. You can describe, refine, and direct images the same way you direct text — with natural language. "Make it more minimal. Change the colour to the palette we settled on. Add a version without the text." Each iteration stays in the same conversation.

Available on Pro plans. The model Claude uses for image generation is tuned for photorealism, illustration, and graphic design outputs — not just abstract art.

Claude can search the web in real time, which means you're not limited to information from its training cutoff. Ask about current events, recent research, live pricing, today's news — Claude searches, reads the sources, and synthesises an answer with citations.

This is different from a standard web search. Claude doesn't just return links — it reads the pages, understands the content, and gives you a synthesised answer. You can follow up conversationally: "What do the critics say about that?", "Are there any counter-arguments?", "Find more recent data."

Claude cites its sources, so you can verify anything it tells you. Particularly useful for research, competitive analysis, fact-checking, and staying current in fast-moving fields.

Claude can read and analyse almost any file you throw at it. PDFs (including scanned documents), Word files, spreadsheets, images, code files, CSVs, presentations. It doesn't just summarise — it understands the content deeply enough to answer specific questions, extract data, compare against other documents, or use it as the basis for new work.

Images: Claude can see and describe images, read text in photos, analyse charts, interpret diagrams, compare multiple images, and reason about what it sees. This includes screenshots, photos, hand-drawn sketches — anything visual.

The combination of file uploads and Projects is particularly powerful: upload your company documents to a Project once, and Claude has that context for every conversation in that project — without re-uploading.

Custom instructions (also called the system prompt) let you tell Claude things about yourself that apply to every conversation — your job, your communication style, what kind of responses you prefer, things to always or never do.

Examples of what to put here: "I'm a solo consultant. Keep responses concise and practical. Don't explain things I obviously know. Use bullet points only when genuinely useful. My preferred response length is medium — not too short, not exhaustive." Or: "I'm not a native English speaker, so avoid idioms. Always give me a summary at the top."

Project-level instructions override global ones — so you can have different "modes" for different projects. The global instructions set your defaults; project instructions specialise Claude for a specific use case.

Claude Code is a command-line tool that brings Claude directly into your development workflow. It's not just code completion — it's a full coding agent that can read your entire codebase, understand its structure, make multi-file edits, run tests, debug, and ship features end to end.

The key difference from Claude in the browser: Claude Code has access to your actual files and can execute code. You can say "refactor the authentication module to use JWT instead of sessions" and it will read the relevant files, plan the changes, make them across multiple files, and verify nothing broke — all autonomously.

This is "vibe coding" at its most powerful — non-developers can describe what they want in plain language and get working code. Developers can offload entire features to Claude while they review and direct.

The Claude desktop app now supports MCP (Model Context Protocol) natively. This means you can connect Claude directly to your tools — GitHub, Slack, Google Drive, databases, Notion, your calendar — without a middle layer like Zapier.

Once connected, Claude can read from and write to those tools inside the conversation. "Check my last 5 GitHub issues and create a priority order." "Look at this week's calendar and draft a focus plan." "Read the latest Slack messages in #marketing and summarise the decisions." Real actions. Real tools. No copy-pasting.

The MCP ecosystem has grown rapidly — thousands of servers exist for almost every major tool. Most are one-line installs. You can find them at mcp.so and the Anthropic MCP registry.