The three, side by side
If you remember one thing: it’s background vs. know-how vs. live access. That tells you which one you actually need.
Projects
A workspace that remembers your files and rules across many chats.
- What it gives Claude
- Static reference material, always loaded.
- Reach for it when
- You keep re-pasting the same docs or context.
- Set it up in
- Projects (left sidebar) → New project
Skills
Reusable instructions that teach Claude a repeatable task or style.
- What it gives Claude
- A procedure it loads only when relevant.
- Reach for it when
- You want consistent output every time.
- Set it up in
- Customize → Skills
Connectors
Links Claude to outside tools so it can read data and take action.
- What it gives Claude
- Live access to a service via MCP.
- Reach for it when
- You want Claude to work in your real tools.
- Set it up in
- “+” in chat → Connectors
Projects
A self-contained workspace with its own chat history, its own uploaded knowledge, and its own custom instructions — so Claude stays on-topic across days or weeks of work.
Worth knowing
- On paid plans, when project knowledge gets large, Claude automatically switches on retrieval (RAG) to expand capacity — you don’t manage this.
- The project name and description are for you, not Claude. The real context has to live in the instructions and the knowledge base.
- Chats inside a project don’t share context with each other — only the shared knowledge base does.
- Archive finished projects to declutter; note that archiving resets sharing to private.
Skills
Folders of instructions (and optionally scripts) that Claude loads only when they’re relevant. They make Claude consistent at specific, repeatable tasks — like formatting a report your way every single time.
1 — Use the built-in skills. Claude already ships with skills for Excel, PowerPoint, Word, and PDF creation. With code execution on (it’s on by default for your Team plan), Claude uses them automatically — ask for “a PowerPoint of Q3 results” and the right skill kicks in. You don’t invoke them by name.
2 — Create your own. A skill is a folder with one SKILL.md file: a short name and description (the description is what tells Claude when to use it), followed by your plain-English instructions.
SKILL.md — or just ask Claude to interview you and write it for you.Worth knowing
- Your Skills list on Team splits into Personal, Shared (sent to you, off until you enable), and Organization (provisioned or published org-wide).
- Shared skills are view-only; if the author updates one, you automatically get the new version.
- There’s a Skills Directory with partner skills (Notion, Figma, Atlassian) built to pair with their connectors.
- Skills follow an open standard (agentskills.io), so a skill you write can work in other tools too.
SKILL.md, and review any skill you download before enabling it. Use a connector for external access, not pasted credentials.Connectors & MCP
MCP (Model Context Protocol) is an open standard for linking Claude to outside tools and data. A connector is the friendly, click-to-set-up version: connect a service and Claude can read from it and act in it, within your permissions.
Worth knowing
- Connectors respect your individual permissions — you can’t reach a teammate’s data through Claude that you couldn’t reach yourself.
- Remote connectors connect from Anthropic’s cloud, not your laptop — so a server behind your VPN or firewall won’t connect unless it’s reachable on the public internet.
- There’s a Connectors Directory of pre-built, reviewed integrations; you can also add custom ones by URL.
- Custom connectors reach services Anthropic hasn’t verified. Only connect servers you trust.
Which one do I reach for?
“Claude keeps forgetting my context.”
Project. Put the docs and standing instructions in one place so every chat starts informed.
“I want the same format / process every time.”
Skill. Write the steps once; Claude applies them consistently whenever the task comes up.
“I want Claude to use my actual tools.”
Connector. Link the service so Claude can read and act on live data — within your permissions.