Most people meet Claude through a single text box. You type a question, you get an answer, you close the tab. That’s maybe a fifth of what the product does now.
Anthropic has spent the last couple of years building a stack around the model: a workspace with persistent context, a coding agent that runs in your terminal, a browser extension, and an open standard for wiring in outside apps. Some of it is genuinely excellent. Some of it is still half-finished. Here’s a working map of the Claude AI tools worth learning, sorted by what each one is actually good at.
The Chat App Is Deeper Than It Looks
Claude.ai is the front door, and most users never walk past the hallway. Two features are buried in the sidebar and deserve a proper look.
Projects keep context so you stop re-explaining yourself
A Project is a folder with a standing brief attached. Drop in your brand guidelines, a style sheet, three example reports, and a paragraph of instructions about how you like feedback delivered. Every conversation inside that Project inherits all of it. For anyone who works on the same client or the same document type week after week, this removes the tedious opening ritual of pasting the same background paragraph for the fortieth time.
The practical limit is worth knowing: each Project has a knowledge capacity, and once you’re past it, Claude retrieves relevant chunks rather than holding everything in view. Curate aggressively. Ten well-chosen files beat eighty that happen to be in the same Drive folder.
Styles fixes the voice problem
You can define a writing style once and apply it to any chat. Feed it three samples of your own prose, describe the register you want, and save it. It’s a small feature with an outsized effect on output that sounds like a person rather than a press release. Pair a Style with a Project and you get consistent voice plus consistent context, which is most of what people are trying to achieve with elaborate custom prompts.
Artifacts Is the Feature Most People Skip
When Claude produces something structural, a panel opens beside the chat and renders it live. This is Artifacts, and it changes the shape of the interaction from reading to reviewing.
Ask for a landing page and you get editable HTML you can preview, tweak, and copy out. Ask for an org chart and you get a diagram you can actually look at. Ask for a budget model and you get a working calculator with sliders instead of a table of numbers you’d have to rebuild in a spreadsheet anyway.
What people build with it in practice:
- Single-page prototypes and component drafts in React or plain HTML
- SVG diagrams, flowcharts, and architecture sketches
- Interactive calculators, checklists, and small tools that run in the preview pane
- Charts and visualisations from pasted data
- Long markdown documents that need real formatting rather than a wall of text
The workflow that gets the most out of it: iterate in the panel, not the chat. Ask for a change to one section, watch it update, keep going. It’s closer to pair-editing than to prompting, and it’s the single biggest reason to use Claude over a plain chatbot for anything visual.
Claude Code for Anyone Who Lives in a Terminal
Claude Code is a command-line agent that reads your repository, plans a change, edits files across multiple locations, runs your tests, and commits. It’s also available as an extension inside VS Code and JetBrains, which is where most people end up using it once the novelty of the terminal wears off.
What separates it from autocomplete-style assistants is scope. You can hand it a ticket that touches eight files and it will trace the call chain before touching anything. It also respects a project-level instruction file, so you can codify rules like “never add a dependency without asking” and have them hold across sessions.
It is not magic. On a large legacy codebase with inconsistent conventions, it will occasionally produce a confident patch that misses why the strange-looking code was strange in the first place. Budget time to review. The teams getting real leverage from it are the ones treating it as a fast junior developer with good recall and no institutional memory, and reviewing accordingly. If you want a sense of how the surrounding field compares, this breakdown of AI tools that beat the generalists at one specific job is a useful sanity check before you standardise on anything.
Connectors and MCP: Claude Reaching Outside the Chat
Model Context Protocol is an open standard Anthropic released for connecting a model to outside systems. In plain terms, it lets Claude read your calendar, search a repository, query a database, or pull a ticket without you copying and pasting anything.
For individual users, this shows up as connectors: Google Drive, Slack, Notion, Linear, GitHub, and a growing directory of others that you authorise once. For teams, MCP is the more interesting part, because it means an internal tool can be exposed to Claude with a small server rather than a bespoke integration.
There are also native integrations outside the chat window. Claude in Slack answers questions about channel history and threads. The Excel add-in builds and audits formulas on live workbooks. The Chrome extension can navigate and read pages on your behalf. All of them are more useful than they sound in a description and less reliable than they sound in a demo. Test on low-stakes work first.
What the Free Tier Actually Gives You
Free accounts get the chat interface, Projects, Artifacts, file uploads, and web search, with usage caps that tighten during busy periods. That’s a genuinely capable package, and for occasional use it may be all you need.
The paid tiers ($20 a month for Pro, more for Max) mostly buy volume and access to the larger models. If you hit a limit twice in a working day, you’re the target customer. If you hit it once a month, you’re paying for headroom you won’t use.
One thing worth doing before you commit a card: understand the trade. Free tiers across the industry differ far more in their limits than in their features, and the honest accounting of what a free AI chatbot actually gives you and what it quietly takes is a better guide to that decision than any feature table. If you’d rather test the water without an account at all, there are also AI tools you can use without signing up, though none of them carry the same context features.
Matching the Tool to the Job
Claude’s stack rewards a bit of deliberate choice. A rough sorting that holds up in practice:
- Writing, editing, analysis: the chat app with a Style and a Project. Skip everything else.
- Anything visual or interactive: Artifacts, iterating in the panel.
- Codebases: Claude Code in your editor, with a project instruction file.
- Recurring team questions: connectors, so Claude reads the source rather than your summary of it.
- Spreadsheet work: the Excel add-in, verified formula by formula.
The comparison that trips people up most often is Claude against ChatGPT, because both are generalists and the differences are in tone, context handling, and how each behaves when a task gets long. There’s a useful side-by-side on which assistant actually fits your work if you’re deciding between the two rather than collecting both.
A final habit that pays off: keep a short list of the three Claude tools you actually use and ignore the rest of the release notes for a quarter. New features arrive constantly, and the temptation to try every one is the main reason people end up with a dozen half-configured setups and no working routine. Pick the two or three that remove real friction, learn them properly, and revisit in three months. If you want a wider shortlist to draw from when you do, this roundup of AI websites worth bookmarking covers the ones that survive contact with actual work.