Claude Quietly Became a Productivity OS. Here's How to Actually Set It Up

Most people using Claude are doing it the hard way. Every conversation starts blank. You re-explain your role, your project, your standards — every single time. You paste the same context into chat after chat, and you scroll through a wall of old conversations trying to find that one answer you know is in there somewhere.

That's not a Claude problem. That's a setup problem. Three features — Projects, Cowork, and Connectors — exist specifically to fix this, and most people never touch any of them. Not because the features are hidden. Because nobody tells you how they fit together.

Here's the mental model. Three layers, each doing a different job:

The mistake almost everyone makes is treating these as three separate features to learn eventually. They're not separate. They're a stack. The Project holds your context. Connectors feed it real data. Cowork does the actual work using both. Skip the Project and Cowork has no standards to follow. Skip Connectors and Cowork is working blind. Set up all three together and every task starts already briefed.

1. Build one Project per work stream — not one Project for everything.

The single most common setup mistake is a single "Work" Project holding marketing files, meeting notes, and client docs all at once. Claude ends up guessing which context applies. Three focused Projects beat one bloated one, every time.

Copy-paste starting instruction for a new Project:

This project is for [specific work stream — e.g. "Q3 client reporting"].
Reference the uploaded files before responding to anything in this project.
Tone: [direct / formal / casual — pick one].
Never suggest a solution without stating the reasoning behind it.
Flag anything you're uncertain about instead of guessing.

2. Keep the knowledge base current, not comprehensive.

Upload what's relevant right now. Replace outdated files instead of piling new versions on top of old ones — a Project with three versions of the same doc confuses Claude more than an empty one. Use clear filenames. When Claude cites a source, you want to know at a glance which one it means.

3. Connect the two or three tools you actually touch daily — not everything in the directory.

The directory lists hundreds of connectors. You need four, maybe five. Start with wherever your decisions actually get made — most people land on some combination of Gmail, Google Drive, Notion, and Slack, because that's where the actual work lives, not because those are the "best" ones. Everything else is noise until you've outgrown the basics.

One accuracy note before you set this up: free accounts are limited to one custom connector. Directory connectors (the pre-built ones for Gmail, Notion, and similar) aren't affected by that limit — it only applies if you're connecting a custom MCP server yourself.

4. Hand Cowork a task, not a chat.

The difference that matters: a chat message gets you an answer. A Cowork task gets you finished work while you do something else. Use it for anything with a clear deliverable — a report, a cleaned-up folder, a first draft — not for back-and-forth thinking.

One distinction worth knowing before you rely on this: Cowork sessions run one of two ways. Remote sessions run on Anthropic's servers and keep working even after you close your laptop. Local sessions — needed for local file access, local connectors, and computer use — require the Claude Desktop app to stay open on your machine while the task runs. If a task needs your local files and you close the app mid-task, it stalls until you reopen it. Plan accordingly if you're kicking off a task and stepping away.

Copy-paste task prompt:

Using the [Project name] project and the [connector] connector,
pull [specific data/files] and produce [specific deliverable].
Deliverable should be [format]. Flag anything missing rather
than filling gaps with assumptions.

5. Chain a Connector into a Project task instead of running them separately.

This is the part almost nobody does, and it's where the real time savings show up.

Copy-paste chained prompt:

Check [connector, e.g. Gmail] for anything from [person/topic]
in the last [timeframe]. Cross-reference against the files in
this project. Summarize what's changed and what needs a response,
in the format this project already uses.

Say you're running a freelance content operation. One Project holds your client's brand voice guide, past deliverables, and style rules. The Gmail connector is attached. Every Monday, a Cowork task runs: check for new client emails, cross-reference against the brand guide, draft responses matching the established voice, flag anything that needs your judgment call. You open your phone Monday morning to drafts waiting, not a blank inbox and a blank page.

That's the entire point of the stack. Not "AI that can do more things" — AI that already knows your context before you ask.

Both show up constantly once you start looking for them:

Three tools. Fifteen minutes to configure properly. The compounding starts on the second use, not the first.

This is Mintd — breakdowns of what's actually worth setting up with AI, not what's trending. Subscribe below and the next one lands in your inbox.