
Cowork is at its best with access to your work and a bit of context about how you do it.
Open the Claude desktop app → at the top, click Cowork. Requires a Pro or Max subscription.
In the sidebar, find Projects and click the “+” button to see the three different ways to create a project: start from scratch, import a project, or use an existing folder.
A Project gives this work a name and a home. The folder you pick is where Claude reads from and where finished work lands. Every task you start inside the Project already knows where to look. To learn more about setting up Projects, visit here.
Claude reads every file inside (PDFs, spreadsheets, Word docs, whatever's there) and saves finished work back to the same place. It reads your old reports to match your formatting, references your templates without being told to.
Open the Customize sidebar → connect and toggle on the tools where your work lives.
Connect tools like Slack, Google Drive, Calendar, Gmail — you do this once, and every task from then on can pull from them.
Instead of copying a Slack thread into your prompt, you mention it ("check what the team said in Slack about this week's earnings") and Claude finds the relevant messages. Claude can also act in connected tools: draft an email in Gmail, save a file to Drive, leave a comment where the discussion lives.
Click into your Project and find the instructions box → write what Claude should know.
A Project has an Instructions that carry into every task you start inside it — Claude reads it alongside everything in your folder.
What goes in is up to you. Standing context you'd otherwise re-explain at the top of every prompt: your role and function, the output format you default to, where things live across tools, judgment calls you want Claude to make the same way every time.
For anything that should evolve — a running log of decisions, notes that grow as the work does — put it in a file in your folder instead. "Add what we covered to my notes file" works on files; the Instructions panel is user-set and stays as you wrote it.
Settings > Cowork → click Edit next to Global Instructions → type and save.
For preferences that don't change between projects: things you want Claude to check before doing, defaults like "save drafts as .docx" or "ask before deleting anything." These carry into every conversation regardless of which Project you're in.
To check your context is being picked up: ask "tell me what you know about how I work here."
Both are optional. With more context, Claude knows more about your work without you spelling it out each time.
Each of these extends Cowork in a different direction, depending on what your work calls for.
Open the Customize sidebar → Plugins → install the one that fits your role.
If you want outputs with domain expertise built in (a finance analyst's framework, a sales team's workflows, a legal reviewer's checklist), install the plugin for your function.
A single prompt can then run a structured workflow with that expertise baked in. You can also customize a pre-built one or build your own for workflows specific to your team. More on plugins →
Type /schedule in any conversation, or go to the Scheduled tab in the sidebar → + New Task.
Once you have a task that works well (a status report, a weekly briefing, a data pull), you can set it to run on a schedule without prompting each time.
Claude walks you through the cadence, the folder, and what the output should look like. As long as the Claude desktop app is open, the task runs on its own and the finished file is waiting when you check. If not, the task will run once you reopen the app. More on scheduled tasks
Open the Customize sidebar → Claude in Chrome → install the extension and enable.
If your work lives behind logins (dashboards, admin panels, web apps), Claude in Chrome lets Cowork see your browser and work inside it.
Cowork can click through pages, fill forms, pull data from anything you're logged into, and make changes in web tools the same way it writes to your folder. Give browser access only to sites you trust. More on Claude in Chrome
Here's how that looks in practice — an analyst updating a research note during earnings season. The shape is the same for any task where you're pulling from files and tools to produce something finished.
A prompt works well when it gives Claude what to look at, what you want back, and where it should go. You don't have to engineer a perfect prompt. Claude will ask follow-up questions for whatever you leave out.
Three of our coverage companies reported earnings this week. The transcripts are in my folder along with our model and last quarter's note. Read the transcripts against our model, check what the team's been saying in #research-desk on Slack, and update the research note. Flag anything that changes our assumptions.
Based on your prompt and what it found, Claude asks a few questions to get the output right — which approach to take, what to prioritize, how the finished work should look.
Pick one of the options Claude offers, or type your own answer. If a question isn't quite right, say so and Claude will ask differently.
A progress panel shows each step — which files Claude's reading, what it's building.
For large tasks, Claude breaks the work into parts and handles them at once: reading several files, pulling from a connected tool, searching the web, all while drafting the output. The kind of multi-source work that would block out your day runs in the background.
Leave it and come back, or type in the chat to redirect if you see Claude heading somewhere you didn't mean. The app stays open; you don't have to be at your desk.
The result lands where you pointed it. In this walkthrough, that's the folder, saved right alongside the transcripts and model.
If the prompt had asked for something different ("draft this as an email in Gmail" or "save to the shared Drive folder"), it would show up there instead.
Summarize something scattered — Claude reads across sources; you get the picture in one place.
"Look through my folder and Slack for everything related to [project]. Give me a one-page summary of where things stand."
Turn a draft into a finished deliverable — the thinking is already in your files — Claude does the shaping.
"Take the notes in this file and turn them into a client-ready memo. Match the format of the last one I sent."
Set up something recurring — Type /schedule and tell Claude what to run, when, and where to save it. The weekly report you draft every Friday becomes a file already waiting when you sit down.