Claude Projects vs Skills vs Connectors: Which to Use When

The three features look similar from the sidebar, and Anthropic has shipped them a year or more apart, so it's easy to pick the wrong one. Projects arrived in June 2024 (Anthropic), skills in October 2025 (Anthropic), and connectors grew from a short list into a large directory. This guide sorts out what each one does, when to reach for it, and how they fit together, using Anthropic's own help pages as the reference.

Projects vs Skills vs Connectors at a Glance

Sources: What are projects, What are skills, Use connectors and the connectors directory docs, all checked November 2, 2026.

What Claude Projects Do

Anthropic describes projects as "self-contained workspaces" with their own chat histories and knowledge bases (Claude Help Center). You upload documents, add project instructions, and every chat you start inside the project sees that material.

A few details worth knowing:

• Everyone can use them. Free accounts can create up to five projects.
• Large knowledge bases are handled. On paid plans, when files approach the context limit, Claude switches to retrieval mode, which Anthropic says expands capacity "by up to 10x."
• Teams can share. On Team and Enterprise, you can give colleagues "Can view" or "Can edit" access.
• A new version is rolling out. Anthropic is testing a redesigned projects experience in beta, starting in Claude Code, with chat, Cowork, Team and Enterprise to follow.

The limit of a project is its boundary. The context is there because you're in the project. Open a new chat outside it and Claude doesn't know any of it.

What Claude Skills Do

Skills are "folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks" (Claude Help Center). The key word is dynamically. Claude reads a short description of each skill, and only loads the full instructions when your request matches. Anthropic calls this progressive disclosure.

There are four kinds you'll run into:

• Anthropic skills for Excel, Word, PowerPoint and PDF files.
• Custom skills you write or upload yourself.
• Organization skills an admin provisions on Team and Enterprise.
• Partner skills in the Skills Directory, from companies such as Notion, Figma and Atlassian.

Skills are available on Free, Pro, Max, Ultra, Team and Enterprise plans, and they need code execution switched on under Settings > Capabilities (Use skills in Claude). You don't need to code to write one. A skill is a SKILL.md file with a name and description at the top and plain instructions below (How to create custom skills). On Mac, Cowork can also record a skill while you walk through a task. Anthropic's own advice is that the best skills solve one specific, repeatable task.

Skills also follow an open standard published at agentskills.io, so the same folder can work in other tools that adopt it. The explainer on agent skills goes deeper, and the list of best Claude skills shows popular ones.

What Claude Connectors Do

"Connectors let Claude access your apps and services, retrieve your data, and take actions within connected services" (Claude Help Center). Think searching Google Drive, reading Slack threads or updating issues in Linear. Claude acts with your permissions, so it only sees what your account can see.

Under the hood, connectors are built on the Model Context Protocol (MCP). There are two kinds:

• Directory connectors, which you browse and add from Claude's catalog. Anthropic's directory docs say the same catalog serves Claude on the web, Desktop, mobile, Claude Code and Cowork, and includes both verified and community entries (Claude docs).
• Custom connectors, which are remote MCP servers you add by URL. They're available on every plan, and Free accounts can add one (Claude Help Center).

Connectors have grown faster than either of the other two features. When Anthropic donated MCP to the Linux Foundation's new Agentic AI Foundation in December 2025, it said Claude had "a directory with over 75 connectors" (Anthropic). On November 2, 2026, Anthropic's connectors page showed 1,051.

Admins have real controls. Team and Enterprise owners decide which connectors are enabled and can set each action to always allow, needs approval or blocked. If you run many connectors, Anthropic suggests switching tool access from Auto to "On demand" once you pass 10 active ones (Claude Help Center). Building your own server is developer work, covered in the guide to creating an MCP server.

Where to Find the Claude Connectors List

There are three places to look:

1. Inside Claude: open Customize > Connectors, or click "+" in a chat, then Connectors, then Manage connectors (Claude Help Center).
2. On the web: claude.com/connectors lists connectors and plugins by category, including productivity, sales and marketing, data and analytics, developer tools, commerce and finance.
3. Your admin's allowlist: on Team and Enterprise, only connectors an owner has enabled will appear for you.

On November 2, 2026, the top of the web list included Google Drive, Gmail, Google Calendar, Canva, Microsoft 365, Notion, Figma, Slack, Atlassian, HubSpot, Asana and Linear, each marked as Anthropic verified (Claude connectors). Ranking in the directory is usage-based, and Anthropic notes that verification isn't a security audit (Claude docs). Treat a listing as a starting point, then check who runs the connector and what it asks permission to do.

Which to Use When: Decision Table

Match the problem you have to the feature that solves it.

A quick test helps when you're unsure. If the answer to "where does this live?" is "in these documents," start with a project. If it's "in how we do it," write a skill. If it's "in another app" or "with someone else," add a connector.

Claude Projects vs Skills

This is the pair people mix up most, because both can hold instructions. Anthropic draws the line clearly: "Projects provide static background knowledge that's always loaded when you start chats within them. Skills provide specialized procedures that activate dynamically when needed and work everywhere across Claude" (Claude Help Center).

In practice:

• Put facts in projects. Product specs, a client's history, a style guide you check against.
• Put methods in skills. "How we write a case study," "how we QA a landing page," "how we format a board update."
• Watch for duplication. If you've pasted the same instructions into three projects, that's a skill waiting to be written.

Custom instructions sit apart from both. Anthropic describes them as preferences that apply broadly to every conversation, while skills are task-specific and load only when relevant (Claude Help Center).

How They Work Together

Anthropic's own framing for skills and connectors is that they're complements: "MCP connections give Claude access to tools, while skills teach Claude how to use those tools effectively" (Claude Help Center).

Here's how the three can stack for a monthly marketing report:

1. Project: "Acme account," with the brand guide, last quarter's reports and the client's goals.
2. Skill: "Monthly report," with your section order, chart rules and the tone you use with this kind of client.
3. Connectors: your analytics data source to pull numbers, Google Drive to save the draft, and Slack to send the summary.

You open a chat in the project and ask for this month's report. Claude has the account context, applies your report method, fetches the data and files the result. Each feature does one job, and none of them has to be stretched to cover the others.

Connectors That Carry Expertise

Most connectors carry data or actions. A newer kind carries know-how: the connector doesn't open a file or post a message, it answers with a method someone has already worked out. That fills a gap the other two features leave. A skill only knows what you wrote into it. A project only knows what you uploaded. Neither helps when the expertise belongs to someone else, or when your own playbook should be kept current in one place instead of copied into every account.

That's what Callstand does. A Callstand stores how you like work done, like your steps, examples and standards, and connects to Claude, ChatGPT and any MCP app. You can also connect a teammate's or an expert's Callstand next to yours. Claude can call it three ways: Plan it returns a prioritized plan, Make it returns the finished piece, and Check it tells you what to fix. Answers come back without revealing the underlying notes, and it says when it doesn't know.

Setup is one link in Claude or ChatGPT; in Claude Code, Cursor or Codex you add the link plus an API key. Pricing, as listed on the same page on November 2, 2026, is $0.50 per call on prepaid credits with no subscription, starting at $5, and calibration with 60 test cases and 10 side-by-side tests costs nothing (Callstand). One limit to know: Callstand's page notes that ChatGPT Plus and Pro are restricted to read-only custom apps.

If your team keeps pasting the same instructions into chats, the post on prompt debt explains why that adds up, and the Callstand guides have more setup walkthroughs.

Frequently asked questions