How to share a Claude Project's context with your team, without a Team plan
On claude.ai, a Project can only be shared between members of the same organization on a Team or Enterprise plan, and only its instructions and knowledge are shared, not the chats. Without those plans, teams copy the instructions and files into a shared folder, a wiki or a repo with a CLAUDE.md, and end up with copies that drift apart. For a single source that every assistant reads and updates, you can also connect the whole team to the same MCP server.
You spent an hour setting up a Project on claude.ai for a client: the brief, the style guide, the meeting notes, carefully tuned instructions. Claude finally answers the way you want. Then a colleague takes over the account and asks: “Can you share it with me?” Depending on your plan, the answer is no. Here is what each tool allows, the workarounds teams use, and what those workarounds cost over time.
What a Claude Project is
A Project is a workspace in claude.ai with its own chat history and knowledge base. You upload documents, write instructions that shape Claude’s answers, and every chat you start inside the Project begins from that context.
According to Anthropic’s help center, Projects are available on every account, free ones included (up to five Projects on the free plan). Paid plans (Pro, Max, Team, Enterprise) add expanded knowledge capacity through retrieval over your documents (RAG).
Who can share a Project, and what gets shared
Sharing is limited to the Team and Enterprise plans, between members of the same organization. On Free, Pro or Max, a Project stays personal.
On Team or Enterprise, the “Share project” button lets you add members one by one, paste a list of emails, or make the Project visible to the whole organization. There are two permission levels:
- Can view: the member sees the instructions and knowledge and can chat in the Project, but can’t change it;
- Can edit: they can also change the instructions, the knowledge and the member list.
A point that is often misunderstood: chats are not shared. Anthropic’s help center says your chats in a shared Project stay private unless you share them yourself. A share link gives a snapshot of the chat at the moment you share it; later messages stay private. Admins can also turn sharing off for the whole organization.
In other words, even on a Team plan, what moves between people is the instructions and the documents. Whatever someone learned or decided while chatting with Claude stays in their own history, unless someone copies it into the Project’s knowledge.
Anthropic is also testing a new version of Projects, in beta in Claude Code (on claude.ai/code and in the desktop app) for the Pro and Max plans. There, Claude runs several tasks in parallel that share instructions, files and a project memory. It doesn’t solve sharing, though: according to the documentation, such a project belongs to a single user and can’t be shared, and this version isn’t offered on Team or Enterprise yet.
The ChatGPT equivalent
OpenAI opened project sharing to all of its plans, free included (Free, Go, Plus, Pro), on top of Business, Enterprise and Edu. The owner invites people by email or by link. There are two access levels: “chat” (see and use the project’s chats, files and instructions) and “edit” (change instructions and files, invite others). Memory in a shared project is limited to that project. The number of collaborators and files depends on the owner’s plan.
| Claude Project | ChatGPT project | |
|---|---|---|
| Can be shared with | Team and Enterprise plans, members of the organization | Every plan, invitation by email or link |
| Instructions and files | Shared | Shared |
| Chats | Private by default, shared as a snapshot | Visible to project members |
| Usable outside the tool | No | No |
Both approaches make sense: Claude keeps each person’s chats private, ChatGPT pools them. Either way, the context lives inside the tool and only helps people who work in that tool.
Common workarounds
Without a Team plan, or when part of the team isn’t on claude.ai, teams improvise. These workarounds do the job, as long as you know where they stop.
Exporting the instructions and files
You copy the instructions into a file, keep the source documents, and each person rebuilds their own Project from that bundle. It’s quick and it works on day one.
The problem shows up the following week: you tweak an instruction in your Project, your colleague adds meeting notes to theirs, and you now have two versions of the context that never come back together.
A shared folder or a wiki
The bundle of instructions and files lives in a shared folder (Drive, SharePoint, a file server) or on a wiki page. That’s better: there is one reference version everyone can read.
But Claude doesn’t read that folder on its own. Each person has to carry updates over to their Project, or paste the content at the start of a chat. And in the other direction, nothing decided in a chat makes it back to the folder unless someone writes it there.
A repo with a CLAUDE.md
For teams that use Claude Code, the CLAUDE.md file at the root of a repository is the built-in mechanism: the Claude Code docs describe it as instructions shared with the team through version control and read at the start of every session. You can split it into files under .claude/rules/, and keep personal preferences in an unversioned CLAUDE.local.md.
It’s a good fit for the stable rules of a code project. For sharing context, it has two limits:
- Claude Code’s auto memory, where it records your corrections and decisions on its own, stays on the machine: the docs state it isn’t shared across machines or in the cloud;
- a Git repo is a poor fit for people who don’t work in a terminal, and it can’t be read from claude.ai.
What these workarounds have in common
They all rely on copies kept up to date by hand. The assistant reads what you give it, but it never updates the shared source. In practice:
- copies drift apart as soon as two people work in parallel;
- a decision made in a chat only exists in the history of the person who made it;
- nobody knows which version is current without checking.
Mixed teams: claude.ai and Claude Code
This is the most common setup in an agency or a small firm: the project manager and the copywriter work in claude.ai, the developer in Claude Code. Each tool has its own mechanism, documented separately: a Project’s instructions and knowledge on one side, the repo’s CLAUDE.md on the other.
The team then has to keep the same context in two places. When the client changes their mind about the tone of the copy, someone has to remember to update both the Project and the CLAUDE.md. Over time, one of them falls behind.
The shared MCP server approach
MCP (Model Context Protocol) lets you connect an assistant to an external tool that provides it with data and actions. What that means for sharing: the context no longer lives inside the assistant, but on a server everyone queries.
On claude.ai and Claude Desktop, Anthropic’s help center says custom connectors using remote MCP are available on every plan, from Free (a single custom connector) to Enterprise. On Team and Enterprise, organization owners add the connector, then each member connects with their own account, which limits Claude to what that person can access.
In Claude Code, an MCP server added with the “project” scope is written to a .mcp.json file at the root of the repo, shared with the team through version control; each person then authenticates, for example over OAuth with the /mcp command.
The result: the project manager on claude.ai and the developer on Claude Code read the same source. More importantly, the assistant can write to that source: record a decision, update a task. Keeping it current no longer depends on everyone’s discipline.
Before picking a server, check a few things: who hosts the data, how you grant or revoke a person’s access, whether the change history is kept, and whether an outsider (a client, a subcontractor) can be given access to a single project only.
Tavimi, one option for this case
Tavimi is the shared working memory for a team and its AI assistants, over MCP. It’s an MCP server with OAuth sign-in, usable from Claude on claude.ai, Claude Desktop, Claude Code and any MCP client.
Each project (a client, a campaign, a job) gets its own space. You invite people by email, and an invited person only sees the space they were invited to, which means you can bring a client in. Notes are written in Markdown and versioned: nothing gets overwritten, nothing gets deleted. They’re filed under categories (decision, pitfall, reference by default) whose writing rules are checked. Tasks carry a status, an owner and a log; when the agent is missing a choice, it asks the question on the task and a human answers from the web app. When you connect an assistant, you tick the spaces it can see. Tavimi is hosted in France.
It isn’t project management software: it sits next to your existing tools, so the team and its assistants start from the same context. The details are on the How it works page, and you can get in touch to try it with your team.
FAQ
What people ask us
Can my teammates see my chats in a shared Claude Project?
No, not by default. Anthropic's help center says your chats stay private even in a shared Project. You can share a link to a chat, which gives a snapshot of it as it was when you shared it.
Can I share a Claude Project with a client outside my organization?
Project sharing, as Anthropic describes it, works with other members of your Team or Enterprise organization. For a client, you either add them to the organization or pass the context another way: a document, a shared folder or a tool connected over MCP.
Do custom MCP connectors work on Claude's free plan?
Yes. According to Anthropic's help center, custom connectors using remote MCP are available on the Free, Pro, Max, Team and Enterprise plans, with a single custom connector on the free plan. On Team and Enterprise, organization owners add them first.
Is ChatGPT easier for sharing a project with several people?
Project sharing is open to every ChatGPT plan, free included, with invitations by email or link, and project members can see the project's chats. That works for a group that does everything in ChatGPT. The context still stays inside ChatGPT, though, and the number of collaborators depends on the owner's plan.
Sources
- What are projects? (aide Claude, Anthropic), accessed October 3, 2026
- Project visibility and sharing (aide Claude, Anthropic), accessed October 3, 2026
- Projects in ChatGPT (aide OpenAI), accessed October 3, 2026
- How Claude remembers your project (documentation Claude Code, Anthropic), accessed October 3, 2026
- Connect Claude Code to tools via MCP (documentation Claude Code, Anthropic), accessed October 3, 2026
- Get started with custom connectors using remote MCP (aide Claude, Anthropic), accessed October 3, 2026
- Let Claude coordinate ongoing work with Projects (documentation Claude Code, Anthropic), accessed October 11, 2026