AI Tools
How to Give Claude Code Memory of Your Business - The CLAUDE.md Method
How to give Claude Code memory of your business: what the CLAUDE.md file actually does, the Standing Brief method for writing it, how to create and update it without touching a terminal, and what memory will not fix.
AI Educator, AI Tools and Training Club · August 24, 2026 · 9 min read
The short version
- Claude Code automatically reads a file named CLAUDE.md in your project folder every time it starts, and treats whatever is written there as standing context - so you stop re-explaining your business at the start of every session.
- Write it as a Standing Brief: who you are and what the business does, the rules it should never break, and the format you want answers in. Ask Claude Code to draft and update the file for you in plain language - you rarely need to hand-edit it.
- Memory is not judgment. A CLAUDE.md file keeps context from session to session. It does not make Claude Code more careful with money, customer data, or anything you would want a human to check first - those rules still belong in the file, written explicitly, not assumed.
The short answer
CLAUDE.md is a plain-text file Claude Code checks at the start of every session inside a project folder. Whatever you write in it - your business, your rules, how you want things done - becomes standing context Claude Code already has, instead of something you retype in every new conversation. Giving Claude Code memory of your business means writing that file once, in plain language, and keeping it current, and you can have Claude Code do the writing and updating for you right inside the app.
What is CLAUDE.md actually?
It is a plain-text file, named exactly CLAUDE.md, that lives in the folder for a project. Every time you start a new session in that project, Claude Code reads it before anything else, so whatever is written there is already known - no pasting, no re-explaining. It is a file, not a setting buried in an account menu, which means you can see exactly what it knows, all at once, in one place.
The problem it solves is the blank-session tax. Without it, every new conversation starts from zero: you re-describe your business, your pricing rules, your tone, your team, before you can even get to the actual task. That repetition is why people quietly stop using AI for anything that requires context - the setup cost eats the time it was supposed to save. CLAUDE.md removes that cost once, permanently, for that project.
What should actually go in it? The Standing Brief
Treat the file as a Standing Brief - the one document you would hand a sharp new hire on their first day, so they never have to ask the basics twice. Three things belong in it: what the business is, the rules it must never break, and the shape you want answers in.
| Part | What goes in it | Example |
|---|---|---|
| Business context | What the business does, who the customer is, the terms specific to how you work | We run a three-person landscaping business. "Job" means one property visit, not one client contract. |
| Hard rules | What Claude Code must never do without asking first | Never send anything to a customer. Never quote a price. Draft only, always. |
| Format preferences | How you want output to look and sound | Reports in a plain table, no headers longer than one line, no corporate language. |
The Standing Brief, in three parts
- Keep it short. A Standing Brief that runs several pages competes with the actual task for Claude Code's attention. A page or less, kept current, beats ten pages written once and forgotten.
- Write it in your own words. This is not a technical spec - it is the same explanation you would give a new hire out loud, just written down.
- Update it the moment a rule changes. An outdated Standing Brief is worse than none, because it looks authoritative while being wrong.
How do I create and update it without touching a terminal?
You do not need to open a code editor or a terminal to write this file. Inside the Claude Code desktop app, tell it directly: "remember that we never quote a price without me checking it first" or "write down that our reports should always use a plain table." Claude Code writes that into the CLAUDE.md file for you, in the project you are working in. The file is plain text, so if you ever want to read or edit it by hand, any text editor opens it - but most owners never need to.
- Open a project in the Claude Code desktop app.
- Tell it the piece of context you keep repeating - your business, a rule, a formatting preference - in the same sentence you would say it out loud.
- Ask it to save that permanently for the project. It updates CLAUDE.md for you.
- Start your next session and confirm it already knows what you told it, before you re-explain anything.
What memory will not fix
A Standing Brief changes what Claude Code already knows. It does not change how carefully it should act, and treating memory as a substitute for judgment is where this goes wrong. If your CLAUDE.md file says "always send the weekly report" without saying who checks it first, you have not saved yourself a step - you have quietly removed the one review that was catching mistakes.
- Never write a rule into memory that grants a blanket permission you would not give out loud - send, delete, spend, or commit to a customer. State the boundary explicitly, in the file, the same way you would state it to a person.
- Memory does not know when something has changed in the real world. A pricing rule from three months ago is still there even after your prices changed, until you go update it.
- A long memory file is not a safer one. Padding it with detail you never check does not add caution - it adds noise the next task has to sort through.
[How to connect Claude to your business tools with MCP](/blog/how-to-connect-claude-to-your-business-tools-with-mcp) covers the same caution from the other direction - what Claude Code is allowed to touch live, rather than what it remembers. The two work together: memory tells it who you are, connectors and permissions tell it what it may actually do.
How I coach members through setting this up
The members who get the most out of this did not sit down and write a perfect Standing Brief on day one. They started with one sentence - the thing they were most tired of repeating - and let the file grow only when something actually came up twice. That is the honest test for whether a piece of context belongs in memory: did you have to say it again.
The mistake I see most often is the opposite: someone pastes an entire onboarding document into CLAUDE.md in one sitting, most of which never gets read again, and then wonders why answers feel generic. A short, current file beats a long, stale one every time.
Frequently asked questions
Does Claude Code remember previous conversations automatically?
Not by default, and that is by design. A CLAUDE.md file is what carries context forward - if you want Claude Code to remember something across sessions, it needs to be written into that file, either by you or by asking Claude Code to save it there for you.
Where does the CLAUDE.md file live?
In the folder for the project you are working in. It travels with that project, so switching to a different project means Claude Code is reading a different file with different context, which is exactly what you want if your businesses or client work do not share the same rules.
Can I keep context that applies across every project, not just one?
Yes. Alongside the project-specific file, Claude Code supports a personal memory that applies more broadly, for the preferences and rules that do not change from project to project. Start with the project-level file - it is the one that matters most for a single business.
Do I need to know how to write in Markdown to use CLAUDE.md?
No. Plain sentences work fine. Ask Claude Code to write and organize the file for you and it will format it. You only need to write Markdown yourself if you choose to hand-edit the file directly.
Is a CLAUDE.md file the same as connecting Claude to my tools with MCP?
No, and the difference matters. CLAUDE.md is static context Claude Code already knows before you ask anything. MCP is live access to a real system it can read from or act inside while it works. Memory tells it who you are. Connectors tell it what it may touch.