How Claude Cowork's Local Folder Access Works

Claude Cowork changes what "working with Claude" means for anything that lives in files.

Instead of pasting content into a chat window and copying results back out, you point Claude at a folder on your machine.

From there, Claude can read what's in it, edit existing files, and create new ones, working through a task across as many files as it needs.

Cowork reached general availability across all paid plans in April 2026, after running as a research preview since January 2026.

This article builds the mental model for what folder access actually grants Claude, and what it doesn't.

Summary

Core Idea: Cowork grants Claude read, edit, and create access to a single folder you explicitly choose, and Claude works inside that folder autonomously rather than waiting for you to paste content in and out.
Why It Matters: Most real work lives in files, not in a chat box, so a tool that can actually touch those files removes a layer of manual copy-and-paste that used to sit between you and getting something done.
Key Concepts: folder grant (the boundary of what Claude can touch), autonomous execution (Claude acting across multiple steps without a prompt for each one), subagent coordination (splitting a task into parallel pieces), Scheduled Tasks (running a Cowork prompt on a recurring cadence).
When to Use: Reach for Cowork when a task requires touching multiple files in a real folder, not when you just want a single answer or a one-off draft.
Limitations / Trade-offs: Claude's access is bounded by the folder you grant, but within that boundary it can read, edit, and create files without pausing for approval on every single action, so the folder you choose is the entire blast radius if something goes wrong.
Related Topics: subagent coordination for parallel work, how Cowork differs from Projects and chat, safety scoping before granting access.

Foundations

The starting point for understanding Cowork is the folder grant itself.

When you set up a Cowork task, you point it at a specific local folder, not your entire computer and not a cloud drive by default.

That folder becomes the entire scope of what Claude can see and touch for that task.

Inside the boundary of the folder you chose, Claude can read existing files to understand what's there, edit files it decides need changes, and create new files as the task requires.

That is a meaningfully different capability from a normal chat conversation, where Claude only ever sees the text you paste in and only ever produces text back for you to handle yourself.

A useful analogy is the difference between describing a repair job over the phone and actually handing someone the keys to the room where the work needs to happen.

In a chat, you describe the file, Claude describes what to change, and you make the edit.

With Cowork, you hand over the room, and Claude does the work directly, inside the boundary you set.

The folder you grant is not a suggestion or a starting point Claude can wander away from.

It is the operating boundary for the entire task, which is also what makes choosing that folder carefully the most consequential decision in setting up a Cowork task.

Mechanics & Interactions

Once a folder is granted, Cowork doesn't just perform one read-and-edit cycle and stop.

It supports multi-step task execution, meaning Claude can plan a sequence of actions, carry out the first ones, check what it produced, and adjust before moving to the next step, all inside a single task run.

That multi-step ability is what makes Cowork suited to work that would otherwise take a long back-and-forth conversation: reorganizing a folder of files, synthesizing information spread across many documents, or applying a consistent change across a whole set of files.

For work that has independent, parallelizable pieces, Cowork can also coordinate subagents, splitting the overall task into pieces that get worked on in parallel rather than one file at a time in sequence.

That matters mechanically because it changes how long a task takes, but it doesn't change the boundary: every subagent working on a piece of the task is still bounded by the same folder grant as the main task.

Cowork tasks can also be packaged into Scheduled Tasks, which let you set up a Cowork prompt to run again on a cadence, hourly, daily, or weekly, with the same folder access and the same access to connected tools and Skills each time it runs.

That recurring capability is powerful precisely because it removes you from the loop on each individual run, which is worth keeping in mind: a Scheduled Task inherits the same folder-access boundary as a one-off Cowork task, just repeated without you watching each time.

You choose a folder
        |
        v
Claude reads what's inside it
        |
        v
Claude plans a sequence of steps (and splits parallel pieces to subagents)
        |
        v
Claude edits and creates files inside that same folder
        |
        v
You review what changed

The diagram above is the core loop for a single Cowork task; a Scheduled Task simply repeats that loop on a timer without a human review step forced between each run.

Advanced Considerations & Applications

The most important advanced consideration is that folder access is an all-or-nothing grant for whatever you point Cowork at.

Claude does not ask permission before every individual edit inside a granted folder the way it might confirm a single action in a more constrained chat tool.

That is the entire point of the autonomy Cowork offers: it lets Claude work through a multi-step task without you approving each micro-step.

But it also means the folder you choose defines the full scope of what could go wrong if a task misfires, whether that's an unwanted edit, an unintended deletion, or a file created somewhere you didn't expect.

The practical implication is that the unit of caution in Cowork is the folder, not the individual file.

Granting access to a narrowly-scoped project folder containing only what a task needs carries a very different risk profile than granting access to a folder that also happens to contain unrelated sensitive material.

Approach	Strength	Weakness	Best Fit
Grant a narrow, task-specific folder	Limits the blast radius to exactly what the task needs	Requires organizing files into that scope first	Any task, and the default recommendation
Grant a broad, general-purpose folder	Convenient, no reorganizing needed	Any misfire can touch unrelated files	Rarely worth it outside quick, low-stakes exploration
Use plain chat instead of Cowork	Zero file-access risk	You do all the file handling yourself	Single-file, low-effort tasks that don't need autonomy

As Cowork has moved from a January 2026 research preview to general availability across all paid plans in April 2026, the underlying mental model has stayed the same even as the surrounding tooling, like Scheduled Tasks and subagent coordination, has matured around it.

The folder grant remains the one concept worth internalizing before anything else, because every other capability, multi-step execution, subagents, recurring schedules, operates inside whatever boundary that grant defines.

Common Misconceptions

"Cowork can see my whole computer once I grant it any folder." Access is scoped to the specific folder you grant for that task, not your filesystem broadly.
"Cowork asks for confirmation before every edit, like a chat suggestion." The autonomy is the feature: within the granted folder, Claude reads, edits, and creates files as part of executing the task, not one approved step at a time.
"A Scheduled Task is safer than a one-off Cowork run because it's routine." A Scheduled Task repeats the exact same folder-access boundary on a timer, with no human review forced between runs, which if anything raises the stakes on getting the initial scope right.
"Subagents mean Cowork can touch other folders in parallel." Subagents split a task into pieces that all still operate inside the single folder grant for that task.
"Cowork is just chat with file upload." Chat with file upload gives Claude the content of files you paste in; Cowork gives Claude the ability to act on files directly, including ones you never manually opened.

FAQs

What exactly does "granting folder access" mean in practice?

You designate one local folder as the scope of a Cowork task.

Claude can read every file inside that folder.
Claude can edit existing files inside that folder.
Claude can create new files inside that folder.
Claude cannot act on files outside that folder for that task.

Does Cowork need my approval before it makes each individual file edit?

No, that's the core difference from a chat suggestion.

Once a folder is granted for a task, Claude works through the multi-step task autonomously, making the edits and creating the files the task requires without a per-step confirmation.

When did Cowork become generally available?

Cowork reached general availability across all paid plans in April 2026, after a research preview period that started in January 2026.

What is subagent coordination, briefly?

For tasks with independent, parallelizable pieces, Cowork can split the work across subagents that work on different pieces at the same time, inside the same folder grant as the main task.

What's a Scheduled Task and how does it relate to folder access?

A Scheduled Task packages a Cowork prompt to run again on a cadence (hourly, daily, or weekly), with the same folder access and the same access to connected tools and Skills on every run.

If I only want Claude to read files, not edit them, does Cowork still work that way?

Cowork's model is built around Claude actively reading, editing, and creating files as needed to complete the task you describe.

If you only want reading and no file changes, a plain chat session where you paste in content is a better fit than granting folder access.

Is the granted folder the only thing that limits what Claude can do?

The folder grant is the primary boundary for file access, but a Cowork task can also have access to connected tools and Skills, which is a separate dimension worth reviewing alongside the folder scope itself.

Can Claude create files anywhere on my computer once I grant a folder?

No, file creation is scoped to the same granted folder as reading and editing; Cowork does not create files outside that boundary.

Why does the size of the granted folder matter so much?

Because the folder is the entire scope of what a task, or a misfire in that task, can touch.

A narrowly-scoped folder containing exactly what the task needs limits how much could go wrong compared to a broad, general-purpose folder.

Is this the same thing as giving Claude "computer use" or full desktop control?

No, Cowork's autonomy is specifically scoped to the granted folder's files, not general control over your operating system or other applications.

How is this different from just uploading files into a chat?

Uploading files gives Claude the content of specific files you chose to paste in.

Cowork gives Claude the ability to read, edit, and create files inside a whole folder directly, including files you never manually opened or uploaded.

Does granting a folder once mean it stays granted for future tasks?

Each Cowork task is set up with its own folder grant, so treat every new task as an opportunity to reconsider whether the scope you're granting is still the right one for what you're about to ask Claude to do.

Claude Cowork Basics - a hands-on walkthrough for granting folder access and running a first task
Understanding Cowork Subagent Coordination for Parallel Tasks - how the parallel-work model builds on this same folder boundary
How Cowork Differs from Claude Projects and Plain Chat - comparing autonomous file access to Claude's other modes
Cowork Safety Checklist Before Granting Folder Access - the pre-flight review for the folder you're about to grant

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.