[FEATURE] Allow Claude to Write/Update Project Files

[SyedMuhamadYasir]

Preflight Checklist

• I have searched existing requests and this feature hasn't been requested yet
• This is a single feature request (not multiple features)

Problem Statement

Summary

Claude can read project files but cannot write, create, or update them. This creates unnecessary friction in workflows where Claude is actively collaborating on a project. Users must manually download files Claude generates and re-upload them to the project — a tedious workaround that breaks flow and defeats the purpose of having persistent project context.

---

The Problem

Current Behavior

Action	Supported?
Read project files	✅ Yes
Create new project files	❌ No
Update existing project files	❌ No
Delete project files	❌ No

User Impact

When collaborating with Claude on a multi-session project:

1. Session continuity is broken — Claude cannot maintain its own memory/checkpoint files in the project
2. Manual file juggling required — Every file Claude generates must be downloaded and manually re-uploaded
3. Context fragmentation — Project state lives partially in Claude's project files and partially in user's local/server files
4. Workflow interruption — Instead of saying "update the checkpoint," users must download → save → upload → confirm

---

Proposed Solution

Proposed Solution

Allow Claude to write to project files with appropriate safeguards:

Option A: Full Write Access (Preferred)

• Claude can create, update, and delete files in the project
• Changes are immediately reflected in the project
• User can review file history/diffs if needed

Option B: Staged Write Access

• Claude proposes changes to project files
• User sees a diff/preview
• User clicks "Accept" to apply changes
• Similar to how code review works

Option C: Designated Writable Folder

• Project has a special folder (e.g., /claude-managed/) that Claude can write to
• Other project files remain read-only
• User explicitly opts in to this folder

---

Benefits

1. True project continuity — Claude can maintain state across sessions without user intervention
2. Reduced friction — No more download/upload cycles for every file change
3. Better collaboration model — Claude becomes a real collaborator, not just a read-only consultant
4. Enables new workflows:
   • Auto-updating documentation
   • Progress tracking files
   • Generated code that stays in project context
   • Test results and logs

---

Security/Safety Considerations

Understandable concerns and mitigations:

Concern	Mitigation
Claude overwrites important files	Version history / undo capability
Runaway file creation	Quota limits on project file count/size
User didn't want the change	Confirmation prompts for destructive operations
Accidental data loss	Soft-delete with recovery period

---

Closing Thoughts

Claude is positioned as an AI collaborator, but the inability to write project files makes it feel more like a read-only advisor. The current limitation forces users into a clunky manual workflow that:

• Wastes time
• Breaks concentration
• Creates version confusion
• Undermines the value of project persistence

For users engaged in multi-session, complex projects — exactly the use case Projects are designed for — this limitation is a significant pain point.

Please consider adding write access to project files.

Written by a frustrated but hopeful user, with help from Claude (who also wishes it could just update the damn file).

Alternative Solutions

N/A

Priority

High - Significant impact on productivity

Feature Category

File operations

Use Case Example

Real-World Example

I'm working with Claude on a multi-day data harvesting project. We created a checkpoint_log.md to track:

• Architecture decisions
• Code changes
• Progress metrics
• Open questions
• Next steps

The ideal workflow:

Me: "Update the checkpoint with today's progress"
Claude: *updates checkpoint_log.md in project*
Done.

The actual workflow:

Me: "Update the checkpoint with today's progress"
Claude: "I can't write to project files. Here's the updated content as a download."
Me: *downloads file*
Me: *opens Claude project settings*
Me: *deletes old checkpoint_log.md*
Me: *uploads new checkpoint_log.md*
Me: "Can you see it now?"
Claude: *reads file* "Yes, I can see it."

This is 5 extra steps for what should be a single operation.

Additional Context

No response

[iBeb]

I just wasted several hours on this exact issue and want to highlight the worst part: Claude doesn't know about this limitation.

I asked Claude to update files in my project. Claude used str_replace and fixed some encoding issues, showed me the changes, and confidently confirmed the edits were complete. When I couldn't see the updates, Claude told me to refresh my browser, blamed caching issues, and gave contradictory explanations about read-only filesystems.

All the work was lost. Encoding fixes across 30+ files, CSV updates, hours of data entry, all gone.

The failure mode isn't "Claude can't do this": it's "Claude thinks it did, tells you it did, shows you output proving it did, then silently fails."

This destroys trust. If Claude is wrong about whether it edited a file, what else is it wrong about?

[p4ck3ts3nd3r]

+1 for this feature - this is exactly what I've been looking for!

I want to strongly support this request. The ability for Claude to edit and create files within the Projects workspace would be transformative for my workflow, and I believe the current architecture already provides the perfect security model for implementing this safely.

Why This Makes Sense from a Security Perspective

The key insight is that Projects already operates as an isolated sandbox. When we upload files to a Project:

The original files remain untouched on our local systems or cloud storage (Google Drive, Dropbox, etc.)
Claude only has access to the copies we've explicitly uploaded to the Project workspace
Any modifications Claude makes would be contained within that workspace
We maintain complete control over what gets synced back to our original files
This is actually a safer model than many other AI integrations that request direct access to cloud storage or local filesystems. The Projects workspace acts as a security boundary - Claude can work freely within it, but the "blast radius" of any mistakes or unwanted changes is completely contained.

My Use Case

I frequently work with documents and files that need iterative refinement. Right now, my workflow looks like:

Upload file to Project
Ask Claude to make specific edits
Copy Claude's suggested changes from the chat
Manually apply them to my local file
Re-upload if I need further iteration
This creates significant friction and breaks the collaborative flow. If Claude could edit files directly in the Project workspace, I could:

Iterate quickly on documents without constant copy-paste
Review Claude's changes and download the modified version when satisfied
Keep my original files safe while experimenting with different approaches
Maintain version history within the Project for comparison
The Staged Approach is Perfect

I especially like the "staged write access" idea mentioned in the original post. Having Claude propose changes that I can review and approve before they're finalized would give me confidence while still dramatically reducing friction compared to the current manual process.

Additional Thought: External Sync Tools

As a future enhancement, this architecture would also enable third-party sync tools to integrate with Projects via API. Users could build automated workflows to sync approved changes back to Google Drive or other storage, while still maintaining the security boundary and explicit approval steps.

Bottom Line

This feature leverages the existing Projects infrastructure and actually enhances security by keeping everything in a controlled workspace. It would make Claude significantly more useful for file-based work without introducing the risks associated with direct filesystem or cloud storage access.

Really hoping to see this implemented - it feels like a natural evolution of what Projects already does so well.

[KirkBrooks]

Agree totally.

Ability to download the Project files should be an option as well.

[syntaxjosie]

Came here looking for this. Yes, please!

[pdonorio]

Iterate quickly on documents without constant copy-paste

omg yes please, it's terrible experience here

[steveylang]

I would love to know if this is even being considered, as it is a major friction point. I have ongoing project files and this would allow me to have the project library version be the master version. Easily referenced in a chat with semantic search capability, accessible across desktop and mobile use, and also accessible to Cowork.

I can accomplish this other by other means, but this would be by far the easiest and most direct way possible.

It also encourages me to stay with a Claude-only workflow, so would be an advantage for Anthropic to offer.

[PsionPhoenix]

I really wish for this feature too. I haven’t been using Claude for long, but having to do changes one at a time, download changed file, delete old file from project, upload new file breaks workflow.
I’ve noticed it more due to writing in python script (new coder, still learning so Claude has been great at explaining things). But it would be nice to have some sort of feature for Claude to update existing files in the project folder. Claude already has a feature to roll back any changes they’ve made if you decide you don’t want them so it wouldn’t be difficult or painful to have this feature.

[addmitt]

Was trying to put together a rather large plan. Thought it would be able to keep a single source of truth doc or .md file in the Project, and continuously edit/make changes. Instantly gave up once I realized it can't update the files on its own.

[steveylang]

One big issue here is that Claude doesn't appear to even have full read-only access to the actual files in the project libraries, only embeddings- which is generally fine but can itself be limiting.

If full read/write access could be added, that would really be the best of all worlds. I'm guessing their recommendation is to just use desktop app file system tools, though that rules out mobile or web clients.

[Haulien]

+1 on this. The core ask is pretty clear: writable project files so Claude can update persistent context across sessions without workarounds. It's a reasonable request given how Projects are positioned as long-running workspaces.

[saif-ellafi]

+1 please

[thorwhalen]

+1 on this.