How to Use Claude Projects: Setup, Limits and What They Cost

If you start every Claude conversation by pasting the same background about your business, your codebase or your dissertation, you are paying a tax you do not have to pay. Claude Projects are the feature built for that problem. A project is a self-contained workspace with its own chat history, its own uploaded files and its own instructions, and every chat you start inside it inherits all three. Learning how to use Claude Projects properly takes about twenty minutes, and it is the highest-return twenty minutes most Claude users never spend.

The catch is that almost every guide on the first page of Google gets the money and the mechanics wrong. Claude Team does not have a five-seat minimum, Pro is not a flat $20, and the single most important thing about a project, that its chats do not share context with each other, is buried or missing entirely. This guide covers the setup, the real limits on each plan, what carries between chats and what does not, how sharing actually works, and the point at which a project stops being the right tool.

The Key Takeaways

• What it is: a workspace with its own chats, its own knowledge base and its own instructions, so you stop re-explaining your context every morning.
• Free tier: Anthropic's help centre states a cap of five projects on Free, while Anthropic's own pricing table lists projects as a Pro feature. The help centre is the operative source.
• Real prices: Pro is $17 a month billed annually, $20 billed monthly. Team is $20 a seat annually, $25 monthly, minimum 2 seats, not the five-seat floor most guides still quote.
• The misconception: chats inside a project do not share context with each other. Only the knowledge base is shared.
• Sharing: Team and Enterprise only, which is the exact opposite of ChatGPT, where sharing reaches every plan including Free.

What Are Claude Projects?

A Claude Project is a self-contained workspace with its own chat history, its own uploaded files and its own instructions. Every chat you start inside it inherits those instructions and can draw on those files, so you stop re-explaining the same background at the top of every conversation.

Anthropic describes them in Anthropic's help centre as workspaces "with their own chat histories and knowledge bases". That phrasing matters, because it is doing two jobs at once. The knowledge base is shared across the whole project. The chat histories are not shared with each other. Almost every problem people have with projects traces back to assuming the second half works like the first.

Three things live inside a project. Project knowledge is the file store: the documents, notes and code you want Claude to be able to draw on. Project instructions are a standing brief that applies to every chat in the project. And the chats themselves are ordinary Claude conversations that happen to start with the first two already loaded.

What You Get on Each Plan

This is where the guides go wrong, so here are the numbers as they read on Anthropic's pricing page and in the help centre today.

The Free Tier Discrepancy Worth Knowing About

Anthropic's two support articles both state plainly that "free users can create a maximum of five projects". Anthropic's public pricing table, meanwhile, marks projects as absent on Free and sells "unlimited projects" as a Pro benefit. Both can be read as consistent, since Free genuinely does not get unlimited projects, but the pricing table will leave a free user thinking they have none at all.

No Anthropic announcement has ever removed projects from the free plan, and independent write-ups quoting the current $17 Pro price still describe the five-project cap. Treat the help centre as the operative source, and if you are on Free, go and look: the button is there. For the full tier breakdown beyond projects, our full Claude pricing breakdown covers what each plan actually buys you.

Does the Five-Project Cap Actually Bite?

For most people, not immediately. Five is enough for one job at work, one side project, one recurring admin task and two spare. The cap starts to hurt when you begin using projects the way they reward being used, which is one per ongoing relationship rather than one per topic. A freelancer with six clients hits it in a month.

How to Use Claude Projects: The Six-Step Setup

The whole thing takes twenty minutes and you only do it once per project.

1. Open claude.ai/projects, or click Projects in the left sidebar.
2. Click + New Project in the upper right corner.
3. Give it a name and a description.
4. Click Set project instructions, write how Claude should behave, and save.
5. Click the + button to upload the documents that carry reusable truth.
6. Star the project so it pins to your sidebar, then start chats from inside it.

Step six is the one people skip, and skipping it quietly wastes the whole setup. A chat started from the sidebar is not in the project and inherits nothing. You have to open the project first and start the conversation from there.

What Belongs in Instructions and What Belongs in Knowledge

Project instructions are for behaviour that never changes: the tone, the format you want answers in, the things Claude should always ask before assuming, the vocabulary your industry uses. They are not the place for this week's assignment. If you find yourself editing the instructions every few days, that content belongs in a chat, not in the standing brief.

Project knowledge is for reusable truth: the brand guide, the product spec, the policies, the price list, the templates you actually reuse. The test is whether you would hand the document to a new hire on day one. If yes, it belongs in project knowledge. If it is a one-off input for a single task, upload it to the individual chat instead and keep the knowledge base lean.

How Big Can Project Knowledge Get?

Anthropic says that "when your project knowledge approaches context limits, Claude seamlessly enables RAG mode to expand capacity by up to 10x". In plain terms, once the material stops fitting in the context window, Claude switches from reading everything to searching it and pulling in the relevant parts. There is no setup on your side.

What Anthropic does not publish is the point at which that switch happens, so do not trust any guide that gives you a precise threshold. Nor does the help centre state a per-file size limit, despite a 30 MB figure being repeated across most of the tool blogs covering this feature. A tight ten-page knowledge base consistently beats a sprawling two-hundred-page one anyway, because retrieval can only surface what you gave it.

What Carries Between Chats, and What Does Not

This is the most useful sentence in Anthropic's documentation and it appears on almost no guide ranking for this feature: "Context is not shared across chats within a project unless the information is added into the project knowledge base."

Read that twice. A project is not one long memory. If you work out a decision in Monday's chat and start a fresh chat on Tuesday, Tuesday's chat knows your instructions and your uploaded files, and knows nothing whatsoever about Monday. The workspace is shared. The conversations are not.

The practical consequence is that anything you want to survive needs to be written down and uploaded. When a chat produces a decision worth keeping, save it as a short note and add it to project knowledge. That one habit is the difference between a project that gets better over months and one that quietly resets every time you open a new tab.

Projects Are Not the Same as Claude's Memory

Claude also has a memory system that carries things about you across conversations, and it is easy to conflate the two. Memory is about you and follows you everywhere. Project knowledge is about a job and stays inside that job. If you want Claude to remember that you prefer British spelling, that is memory. If you want it to know your company's refund policy, that is project knowledge. Our guide to Claude's memory covers the distinction and the controls in more depth.

Projects Are Cheaper Against Your Usage Limits

One benefit nobody advertises: uploading to a project is more efficient than pasting the same document into chat after chat. According to Anthropic's usage-limit guidance, documents uploaded to a project are cached, and "every time you reference that content, only new/uncached portions count against your limits". If you keep running into caps, this is a real lever. Our breakdown of Claude's usage limits explains what you are actually working against.

How to Move an Existing Chat Into a Project

You do not have to start over. Click the dropdown arrow next to a chat's name, choose Add to project, then browse for the project you want. If you have a backlog, the chat history page lets you move several at once rather than doing them one by one.

Moving a chat in does not retroactively give the rest of the project access to what was said in it. The rule from the previous section still holds: only project knowledge is shared. Moving a chat is filing, not merging.

How to Use Claude Projects With a Team

Sharing a project requires a Team or Enterprise plan. There is no version of this on Free, Pro or Max, and that surprises people who assume a $100 Max subscription outranks a $20 team seat. It does not, because sharing is a workspace capability rather than a usage tier.

On Team and Enterprise you can share with named individuals, add a list of email addresses in bulk, or open a project to the whole organisation unless an admin has disabled that. Permissions come in two levels, Can view for read-only access and Can edit for full modification rights. Team starts at $20 a seat billed annually with a minimum of two seats, which means the real floor is $40 a month. Several guides ranking for this topic still quote a five-seat minimum and a $125 floor, which is where that number came from: the minimum used to be five and came down to two during 2026.

If you are on Pro and need a colleague to see the work, the honest answer is that you export or paste it. There is no supported workaround, and treating a shared login as one is a security problem rather than a solution.

Claude Projects vs ChatGPT Projects

Both products shipped a feature with the same name and then made almost opposite choices about who gets it. If you are weighing the two, or moving from ChatGPT to Claude, this is the table that matters.

The short version: Claude is more generous with what a project can hold and stingier about who can see it, and ChatGPT Projects work the other way round. If collaboration is the point, ChatGPT wins outright below the business tiers. If the point is one person working deeply against a large body of material, Claude's retrieval behaviour is the better fit.

When Projects Go Wrong

Four failures account for nearly every complaint, and all four are fixable.

Claude ignores a file you uploaded

Usually the knowledge base has grown past the context window and retrieval is now selecting what to read. Ask a question that names the document explicitly, or trim the knowledge base back to the material that carries reusable truth. Vague questions get vague retrieval.

A new chat has forgotten an old decision

Working as designed, and the single most common surprise. Context does not cross between chats. Write the decision into a short note, upload it to project knowledge, and it will be there next time.

Answers turned generic as the project grew

A knowledge base full of near-duplicates gives retrieval too many plausible matches. Delete superseded versions rather than adding new ones alongside them. One current document beats four historical ones.

A colleague cannot see the project

Check the plan before debugging anything else. Sharing does not exist below Team, so on Pro this is not a permissions bug, it is the product.

Where Claude Projects Stop Helping

Projects are excellent at one thing: keeping a long-running job's context in one place so you stop rebuilding it. They are poor at anything that needs conversations to talk to each other, because they were not built for that. They are also poor at collaboration below Team, and they are no help at all if your work is spread across several places at once.

They are also Claude-shaped and Claude-only. If your habit is to draft in one model and sanity-check the answer in another, a per-vendor workspace does not travel with you, and you end up maintaining the same brief in two or three places. That is worth knowing before you invest a weekend in building out fifteen projects. A Claude desktop client for macOS keeps the app a keystroke away, which takes some of the friction out either way.

Should You Use Claude Projects?

Yes, if you have any job you come back to more than twice a week. The setup cost is twenty minutes and the return is not having to re-explain yourself, which for most people is the single largest source of wasted time in an AI tool.

Build one project first, not fifteen. Pick the job you explain most often, write the instructions for behaviour rather than for this week's task, upload three or four documents you would hand a new hire, and then make the habit of writing decisions back into project knowledge. If you are on Free, you have five slots to prove the idea works before the cap is a reason to pay for anything.

Frequently Asked Questions