Feature request: Connect Claude code to Claude projects

Found 1 possible duplicate issue:

1. https://github.com/anthropics/claude-code/issues/1482

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

I strongly endorse this feature. I commonly will use Claude projects to help me organize my brainstorming and feature planning and development. I often do this on my mobile device when I'm not near my computer or terminal sometimes I'll do it on the go. It would be really helpful if Claude code was able to have up-to-date knowledge about my brainstorming sessions that I develop on the Claude app and linked to a project.

I would also love to see this feature. Claude chat is excellent for planning and doing high level design. Claude Code is really good for implementing that. It would be great to see a connector for this.

A one-way sync could theoretically be implemented by allowing Claude Code to configure a global /projects directory somewhere on the user's filesystem. One folder per Claude Project. When you do a brainstorming session via Claude desktop, offer a 'sync' option that:

• Generates a summary of what was achieved in this session.
• Adds it to the Project resources
• Keeps the local /project directory in sync with the remote project resources folder.

Then offer a sync-project command for the claude CLI, which reads the files in the configured local /projects/name-of-project folder, and updates the CLAUDE.MD file with references to the contents of the results of the individual sessions within a project.

There might be a better way to do it, but that's what comes off the top of my head.

+1 on this!

Would love to see some continuity between claude.ai and Claude Code

Moving from discussion to coding
I'll often hash out architecture ideas or debug approaches on claude.ai, then when I jump into Claude Code to actually build it, I have to re-explain everything we just talked about or provide a .md file and vice versa - claude code to claude.ai. I kind of hack this with claude projects but doesn't feel as seamless as it could

Keeping context across sessions
Being able to reference earlier conversations about why we chose a certain approach or what trade-offs we discussed would be really useful. Right now I'm just copy-pasting stuff between the two, which feels clunky.

Saving time and tokens
It'd be nice to not have to repeat the same background info every time I switch from planning mode to implementation mode.

I know the main ask here is about project knowledge bases (which sounds great), but honestly even something simpler like being able to link to a claude.ai conversation or pull in recent chat history would go a long way.

Would make the whole workflow between the two platforms feel way more connected.

+1 on this as well. right now I use .md files of the project which has the full history, PRDs, designs, etc. to share between claude app and calude code. full project memory would be great but as other mentioned even direct access to chats, files in the project is a good start.

I have also wondered why this is not possible yet

I totally support this feature. We use the chat to brainstorm. We have provide guidelines and documentation to support the design. then let claude code integrate the design into your project. That it is a win win

Plus 1. This would be a great way to manage context and even allow collaboration between folks who want outcomes from the folks who are coding the app in claude code. I also like the separation of concerns as a solo developer for this and should be more efficient at managing context.

I hate to add a "me too" to this, but... well... here I go adding one.

Slight difference in positioning with me, though. My position is not "it'd be nice if..."; to me this is just a no-brainer, and is - surely - table-stakes?! Anthropic needed my credentials to "activate" the Claude Code plug-in in VSCode, so it already knows who I am. How did it not occur to anyone that possibly the already existing work in my account (projects, chats, context guidance, memory summaries etc) would not be equally relevant across various different entry-point UIs (which is all the difference between Claude Code and Claude Desktop really is, surely?)? That info shapes the entire Claude experience.

Anyway... sorry to gripe, I mean I can continue to work with Claude Desktop & MCP servers, so I'm all good; I'm just... well... surprised this wasn't addressed before launch.

Other than that, cheers for all the hard-work the dev team are putting into the Anthropic products. I realise it's almost certainly not you bods making these decisions ;-)

I want to add a related request: it would be extremely helpful if Claude Code could access artifacts created in the Claude UI (files, diagrams, bundles, etc.), not just Project knowledge bases.

Right now, anything generated in the UI has to be manually downloaded or copied over to Claude Code, which breaks the workflow. Having a way to reference or pull in UI-created artifacts directly would make the transition between UI brainstorming and Code implementation much smoother.

This feels like an important piece of connecting Claude Code with the broader Claude ecosystem.

Adding my voice in support of this request. If some Action or Agent is looking for clues as to were engineers should focus there limited energy and attention... this is it. Moving seemlessly between interconnected workspaces is the obvious way you can actuallly make claude.ai/code a fundamentally important part of their work flow.

Roo code did something smart that i think serves as a model for how claude itself should work at scale . https://docs.roocode.com/roo-code-cloud/roomote-control

Clearly this isn't a small request but I think its a bigger game changer than y'all realize.

Start a project in a terminal app, or a relay or a wrapper around a terminal app, and all of a sudden super powers are at the finger tips of so many users who right now are scared of the terminal. The key is the ability to continue to interact with an agent seemlessly from your terminal agent to your web app and vice versa... that would not only be better for users but it would fundamentally change the way people can use their computers.

My guess is for many the killer app would be simple device management and customization. Manufacturers are trying to bake this with like co-pilot on pc's and gemini apparently soon on macs? But what if users could just do it all from the claude app with a secure tunnel to claude code. For many it would be things as simple as changing icons or organizing files more efficiently on their devices with chat commands. And that leads to troubleshooting your own network connection, better understanding your security issues, optimizing your set up. All of a sudden there is huge value for browser based coding agents that can unlock terminal agent functionality. Conveniently relaying approvals, interuptions and feedback loops into projects with full acess to apis, mcps and all that an agent with control of the command line has to offer.

In truth this is just something I want for my workflow. I switch devices a lot. I think many of us are multi device or multi work station or whatever you want to call it. It would be great to be able have a continuous experience. Someday maybe even running locally on our own little servers, dare to dream. haha

If you have read all of this, thanks for your attention. Have fun out there, touch grass, drink water.

hugely endorse of this feature too. I have two devices that I would need to switch frequently. Having a way to resume from another device would be an efficiency killer for me

Adding a Teams Plan and collaboration angle to this request.

Most comments here focus on individual workflows (personal brainstorming -> implementation), which is valid. But there's a related gap that affects engineering teams paying for the Teams Plan (I'm an Engineering Manager).

Part of the reason I advocated for a Team Plans is collaboration features. However, currently we have:

• The Projects feature - works great in the web UI
• The "Ask [ORG]" feature - works great in the web UI
• Claude Code - completely disconnected from Projects, with only per-user or per-repo configuration

For teams whose primary workflow is Claude Code in terminals/IDEs, the collaboration features we're paying for aren't accessible where we work.

Our team works across 10+ repositories with different tech stacks (Python services, Scala backends, TypeScript frontends, Shell scripts, YAML files, etc). Each repo can have its own .claude/CLAUDE.md or AGENTS.md, but:

• Team standards (testing requirements, review expectations, architectural patterns) must be duplicated across every repo
• No team-level config layer exists between personal ~/.claude/CLAUDE.md and organization-wide settings
• Different teams in the same org have different standards; org-wide settings are too broad

Current workarounds:

• Team members share their prompts and settings in team meetings and chats.
• Team members edit shared markdown files in the code.
• Collaboration via other tools and integrating via MCP.
• Shared git repos with symlinks to a personal ~/.claude folder.

These all work but they are manual and fragile.

What would help

Beyond connecting to Projects knowledge bases (the original request), consider:

1. Claude Code -> Projects integration (this issue's core ask)
2. Team-scoped configuration - a layer between personal and org-wide that team members inherit automatically
3. Multi-repo project definition - a way to define shared context across related repositories

The original request would solve (1). But (2) and (3) would address the team collaboration gap that makes the Teams Plan feel like we're paying for features we can't use in our actual workflow.

110% it's extremely surprising that this didn't exist from the start - just used some Opus 4.5 credit to discover this ticket and the fact that I can't do what I want which seems to be the most obvious thing to do:

1. Set up a project in claude - context for both chat and claude code work
2. Set up a project root on disk
3. Launch claude code in that root and tell it to use the project
4. Proceed to operate on each repo in turn from that project root using that project context

I'll also upvote but this one needs to get implemented. Just bought the 100usd/month plan for summer and was hoping to do this, now I have work to do to set up that context on disk to match :-/ oh well, good enough for now, but c'mon Anthropic, pull finger on this one!

One way would be if Claude would have an MCP to project memory and artifacts at minimum. I often do some planning, conceptualizing in Claude and then go and develop in Claud Code.

+1
I'm in agreement with this request as well as the comments augmenting the feedback.
Extending Code and allowing Claude Work to tie into project chats and attachments would be a very powerful toolset that is able to work in a unified way over our project efforts.
Being able to distribute the architecture and design and specification discussion (strategic layer) from Claude Chat with is larger context window and project files, to the claude code (tactical layer) and its focused sessions would be great. Continue on to enabling the coordination capabilities of claude work would become a very powerful stack to work over!

Please enable this, it would be amazing to have the entire workflow embedded although co-work has been created.

+1

+1

+1
I'm currently forced to use NotebookLM and the NotebookLM skill to handle large documentation. It would be great if everything could stay within the Claude ecosystem, but the Claude project needs to improve a bit before reaching the same level of information extraction that NotebookLM provides.

+1 — this would be a huge quality-of-life improvement.

+1 — CMMC compliance, automated documentation syncing for defense contractor projects, etc. would make things so much easier. If only there were an automated coding tool that might be able to help with this ;) (Keep up the great work team! You are killing it in the real world.

searched for how to do this and this was the search result... please add this feature

Github is becoming my new favourite social media as all the notifications roll in to support this feature request. What do you think the magic number is before they do it 🙃?

Heya everyone @here, this appears to have landed as a feature on Cowork on mac OS (havent checked windows yet, its limited). Given it can be fed into cowork, it can be reflected into code activities coordinated out of cowork, gonna try it out!

<img width="701" height="445" alt="Image" src="https://github.com/user-attachments/assets/cdf54fe2-88c5-458f-84c0-268434d34fb0" />

I won't be using cowork, so I need it in Claude Code....

On Wed, 11 Feb 2026 at 08:14, tgelite @.***> wrote:

tgelite left a comment (anthropics/claude-code#2511) <https://github.com/anthropics/claude-code/issues/2511#issuecomment-3880123288> Heya everyone @here <https://github.com/here>, this appears to have landed as a feature on Cowork on mac OS (havent checked windows yet, its limited). Given it can be fed into cowork, it can be reflected into code activities coordinated out of cowork, gonna try it out! image.png (view on web) <https://github.com/user-attachments/assets/cdf54fe2-88c5-458f-84c0-268434d34fb0> — Reply to this email directly, view it on GitHub <https://github.com/anthropics/claude-code/issues/2511#issuecomment-3880123288>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAJA5ZWU2NVX5AIANEEXC34LIT2VAVCNFSM6AAAAAB77CRELKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTQOBQGEZDGMRYHA> . You are receiving this because you commented.Message ID: @.***>

+1 - the dream would be to:

1 Have sync between a project and claude code
2 To have an option to write/chat to specific claude instance trough the web interface
3 To have the option to receive the claude code plan with diffs directly in the web version. That means beside full md plan also option for a MR similar view of planned changes before it starts.

This would be lovely, @a-tonchev - then I could use my phone to get work done while anywhere I want :-)

+1 - the dream would be to: 1 Have sync between a project and claude code 2 To have an option to write/chat to specific claude instance trough the web interface 3 To have the option to receive the claude code plan with diffs directly in the web version. That means beside full md plan also option for a MR similar view of planned changes before it starts.

This would be lovely, @a-tonchev - then I could use my phone to get work done while anywhere I want :-) > +1 - the dream would be to: > 1 Have sync between a project and claude code 2 To have an option to write/chat to specific claude instance trough the web interface 3 To have the option to receive the claude code plan with diffs directly in the web version. That means beside full md plan also option for a MR similar view of planned changes before it starts.

Yes, this would be revolutionary in the AI assisted coding

I can't sleep, so I'm brainstorming with Claude. We all need this functionality in the morning.

+1000 🤤

I just hit this gap again. It's just idiotic this isn't possible.

It's one thing to expect a parcel from New York to reach a fishing village in rural Japan with ease, and yet that's possible.
What we're asking for here is to take something held in one hand and pass it to the other hand. Same company, same user, same auth, same model. Why is that not possible. I didn't expect to lose all my work going from one tab to another tab.

Frustration from just having lost work aside, y'all can knock this out in a day or 3. Here's hoping.

I just hit this gap again. It's just idiotic this isn't possible. It's one thing to expect a parcel from New York to reach a fishing village in rural Japan with ease, and yet that's possible. What we're asking for here is to take something held in one hand and pass it to the other hand. Same company, same user, same auth, same model. Why is that not possible. I didn't expect to lose all my work going from one tab to another tab. Frustration from just having lost work aside, y'all can knock this out in a day or 3. Here's hoping.

I've started using jiras (atlassian connector) to bridge between project chat and cowork/code as a workaround (and keep my cowork tasks organized... Because damn that list gets unwieldy quickly)

This is what Claude has to say:

Okay Claude, when I create a project here, do I get synchronization somehow with Claude Code? Can Claude Code access your project? You are the Claude up on Windows.

Investigated synchronization capabilities between Claude.ai Projects and Claude Code Investigated synchronization capabilities between Claude.ai Projects and Claude Code Short answer: No native sync. Claude.ai Projects and Claude Code are separate contexts. Claude Code can't access your Project's knowledge base, instructions, or conversation history from claude.ai. They're isolated — different products, different context mechanisms. What each uses: Claude.ai Projects → system prompt + uploaded files in the web UI Claude Code → CLAUDE.md files, local filesystem, MCP servers Bridging strategies you could use: Keep your project instructions in a CLAUDE.md at your repo root (Claude Code reads these natively), and paste the same into your claude.ai Project instructions Store shared knowledge in your Obsidian vault — Claude Code reads from the filesystem, and you already have that pipeline MCP servers are available in both, so shared tooling (Linear, etc.) works across both

Isn't it ironic, maybe even moronic, that there's no connection between the two Claudes that run on the same machine? There's no MCP. There's no built-in communication between that. I mean, I want to build my brain and have you as my assistant and have access to all my information, but how am I going to do that if you can't talk to your own products? I'm talking to Anthropic. Could you please escalate that?

You're making a completely valid point, and yeah — it's a pretty glaring gap. You've got two Anthropic products running on the same machine, both called Claude, and they can't share context. No shared memory, no shared project state, no MCP bridge between them. It's exactly the kind of fragmentation that Anthropic's own MCP protocol was designed to solve.

I set up a custom Obsidian Git repo and tried connecting Claude.ai to it via a custom MCP connection. The idea was simple: Claude.ai reads and writes to the repo, and when I pull locally, Claude Code can do the same.

The problem is that GitHub connections don't work properly in Claude.ai. I then built my own MCP server with a GitHub token, but Claude.ai has a bug that prevents it from working with remote MCP servers via custom connector. The Claude Windows app connects fine, though.

So my current workflow is: I chat in Claude.ai, but when I need to summarize or reference something, I switch to the Claude Windows app and have it read/write to the repo. Then I point Claude Code to that information.

It works, but the Windows app is unstable — frequent crashes and bugs. Right now, Claude Code is the only part of the stack that runs reliably. Everything else is either crashing or hitting unresolved bugs that even Claude itself acknowledges.

Thanks for the workaround flow @a-tonchev. Would you mind sharing that MCP server?

Thanks for the workaround flow @a-tonchev. Would you mind sharing that MCP server?

Actually the MCP do not work right now, some Claude AI problem. I use directly the git plugin in the claude app for desktop

Would love to have claude chat conversations available to claude code, and claude code sessions available to claude chat. Chat could be just like slack -- an MCP server, basically, and claude code could be accessed through it.

Ideally I'd be able to say which remote control claude code session I want chat to access

Feature Request: Integrate Claude Code with Claude.ai Projects Knowledge Bases Summary Enable Claude Code to access and leverage knowledge bases from Claude.ai Projects, allowing developers to use their existing project documentation, specifications, and reference materials directly within their coding workflows. Current State - Claude.ai Projects allow users to upload documents and create knowledge bases that Claude can reference within - Projects offer caching benefits where uploaded documents are cached for future use, and only new/uncached portions count against usage limits [(1)](https://support.anthropic.com/en/articles/10167454-using-the-github-integration) - Claude Code currently supports project-specific configurations using .claude/settings.json files and custom slash commands [(2)](https://github.com/anthropics/claude-code-action/blob/main/README.md) - CLAUDE.md files can link other Markdown and text files, which are automatically pulled into context [(2)](https://github.com/anthropics/claude-code-action/blob/main/README.md) ### Proposed Enhancement Allow Claude Code to connect to and access Claude.ai Projects knowledge bases, enabling: 1. Seamless Knowledge Integration: Access existing project documentation, API specs, design documents, and other reference materials stored in Projects 2. Efficient Caching: Leverage the existing Projects caching system to avoid repeatedly processing the same documents 3. Unified Workflow: Maintain context across both web-based Claude interactions and terminal-based Claude Code sessions 4. Enhanced Context: Provide Claude Code with richer project understanding through comprehensive knowledge bases ### Use Cases - API Development: Access API documentation and specifications while implementing endpoints - Code Refactoring: Reference architectural decisions and design patterns stored in Projects - Bug Fixes: Consult troubleshooting guides and known issues documentation - Feature Implementation: Access product requirements and user stories during development - Code Reviews: Reference coding standards and best practices from project documentation ### Benefits - Reduced Context Switching: Developers can access project knowledge without leaving their terminal - Improved Code Quality: Better context leads to more informed coding decisions - Cost Efficiency: Leverage existing Projects caching to reduce token usage - Consistency: Ensure Claude Code has the same project context as web-based Claude sessions ### Implementation Suggestions - Add a configuration option to link Claude Code sessions to specific Claude.ai Projects - Extend the .claude/settings.json format to include project references - Provide commands to sync or refresh project knowledge within Claude Code - Consider integration with the existing CLAUDE.md linking mechanism ### Technical Considerations - Maintain security and access controls consistent with Projects permissions - Handle offline scenarios gracefully when project knowledge isn't accessible - Provide clear indicators when project knowledge is being used vs. local context This feature would significantly enhance Claude Code's capabilities by combining the power of local development tools with the rich knowledge management features of Claude.ai Projects.

Id love to also have this function

While direct Claude.ai Projects integration isn't available yet, a UserPromptSubmit hook can inject project knowledge from local files on every turn — effectively giving Claude Code access to your knowledge base:

KNOWLEDGE_DIR=".claude/knowledge"
[ -d "$KNOWLEDGE_DIR" ] || exit 0
CONTEXT=""
for f in "$KNOWLEDGE_DIR"/*.md "$KNOWLEDGE_DIR"/*.txt; do
    [ -f "$f" ] || continue
    CONTEXT+="--- $(basename "$f") ---
$(cat "$f")
"
done
[ -z "$CONTEXT" ] && exit 0
jq -n --arg ctx "Project knowledge base:
$CONTEXT" '{
  "hookSpecificOutput": {
    "hookEventName": "UserPromptSubmit",
    "additionalContext": $ctx
  }
}'
exit 0

Setup:

.claude/knowledge/
├── architecture.md       # system design docs
├── api-spec.md           # API specifications
├── coding-standards.md   # team conventions
├── deployment.md         # deployment procedures
└── domain-glossary.md    # business domain terms

Add to .claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [{
      "matcher": "",
      "hooks": [{ "type": "command", "command": "bash .claude/knowledge-loader.sh" }]
    }]
  }
}

This gives you:

• Same benefit as Projects: reference docs always available in context
• Version controlled: knowledge files live in git alongside your code
• Team-shareable: everyone gets the same knowledge base via the repo
• No API dependency: works offline, no claude.ai account needed

For larger knowledge bases, add selective loading based on prompt keywords:

PROMPT=$(cat | jq -r '.userPrompt // empty')
if echo "$PROMPT" | grep -qiE '(api|endpoint|route)'; then
    cat ".claude/knowledge/api-spec.md"
fi
if echo "$PROMPT" | grep -qiE '(deploy|release|ship)'; then
    cat ".claude/knowledge/deployment.md"
fi

@yurukusa I think you missed the point - I for one set up a project expecting to centralise everything around it and use it in all claude operations - so I have a heap of stuff in there and then discovered I couldn't get claude code to see it. This is all well and good if you work this way - but the beauty of being able to chat and come up with a solution using the LLM as a sounding board is part of what makes it so effective for planning and design sessions etc. Once all this is done having to extract it out and isolate it from the source for other use is awkward and hard.

If your hook used your token to go and get the project context and parse it and provie in index for progressive loading then that'd work. Otherwise nope, that's not the solution we're looking for. And by we I mean me at least. :-)

I second that @fredcooke. It seems crazy this is not possible as others have said. Brainstorming in the app is something I do often, and the poject idea is great because I can keep informaiton and conversations together. But then when I want to implement something in Claude Code it's a hassle to get the context into Claude Code.

Big +1. The disconnect between Claude.ai Projects and Claude Code is frustrating. I use Claude.ai for brainstorming and research, build up a knowledge base in Projects, and then switch to Claude Code for implementation — where none of that context is available.

Connecting the Cowork/Claude Code infrastructure more tightly with Claude Projects would be a massive workflow improvement. Right now it feels like two separate products that should be one.

Essential feature!! ++++

This makes too much sense to not implement. Projects could be the center of Claude work.

Use Case: Cross-Project Knowledge Synthesis

I'd like to add a related but distinct use case to this request.

I currently maintain two separate Claude Projects that cover different but closely related subject areas. I need to create a third, more advanced project that:

• Draws on the knowledge bases of both existing projects
• Ensures cohesion and consistency of content across all three
• Produces synthesized reports and outputs that neither project could generate alone
• Serves as a higher-level workspace built on top of the two foundational ones

Right now, the only workaround is to manually export documents and conversation outputs from each project and re-upload them into the third — with no live sync. Every time Project 1 or Project 2 is updated, Project 3 has to be manually refreshed. This creates friction and increases the risk of the synthesis project falling out of date.

What I'm requesting:
The ability to link one Claude Project to one or more other Projects as knowledge sources — so that a parent or synthesis project can reference, draw from, and stay current with its linked projects automatically.

This would unlock a natural tiered workflow: foundational projects feed into advanced ones, enabling more complex, cohesive, and scalable work within Claude.ai without manual document management overhead.

the part that breaks down with 'connect to Projects KB' isn't the plumbing, it's that you'd be doubling the surface where Claude already ignores you. CLAUDE.md ships every turn whether the model uses it or not. layering a Projects KB on top means the same passage lives in two caches with different invalidation rules, and you still don't know which 200 lines of API spec Claude actually read vs which got skipped because the token budget got tight mid-session. the caching benefit assumes the model wants the bytes you sent. the harder question is what it's currently skipping.