Why Claude memory does not reach your teammates
Claude's built-in memory learns from your chats and remembers your preferences, but it belongs to your account. Your teammate's Claude cannot see it, and no plan merges two people's memories into one. No plan merges two people's memories into one shared brain. The gap is not the tier you pay for; it is where the shared context lives.
Claude Projects help a little: a team can share files and instructions inside one project. But that is a fixed brief people load, not a live memory that each session updates for the next person. So claude team memory is really a question about where shared context lives — not which plan you pay for.
The pattern: one store every Claude reads and writes
The fix is to keep the team's project memory in one place that lives outside any single account, then connect each teammate's Claude to it over MCP (Model Context Protocol). Every session now starts by reading the same notes, and can write new ones back.
- One shared source. Decisions, names, gotchas, and next steps live in one spot, not scattered across everyone's private chats.
- Read on the way in. A teammate's Claude pulls the current context before it answers.
- Write on the way out. When something is decided, Claude records it for whoever comes next.
The write-back loop
This is the part per-user memory cannot do across people. On Monday your Claude records a decision — "we moved auth to the new provider; leave the old callback alone" — and writes it to the shared store. On Wednesday a teammate asks their own Claude about auth. It reads your note and answers with it, and you were never in the room.
The memory compounds. Each session leaves the next one better informed, no matter who is driving. One person's Claude teaches the whole team's Claude, which is exactly what a single per-user memory can never do.
Where agent-wiki fits
agent-wiki is one honest way to run this pattern. It is a plain-markdown wiki your team shares, and every connected assistant can both read it and update it — not a read-only knowledge panel. Because it is neutral, the same wiki answers to whichever tool a teammate opened: Claude, Claude Code, Cursor, or ChatGPT through its connector.
Two details make a shared memory safe to trust. Writes use optimistic concurrency — each edit carries the version it was based on — so two people's assistants never silently overwrite each other. And every read and write is logged with full version history, so the memory stays inspectable rather than mysterious; see security & trust for how that works.
If you want the longer story of why this problem exists, why agent-wiki exists covers it. For the full feature list and current plans, the home page has the detail.
Get your team on one memory
You do not need an Enterprise plan to give a team shared Claude memory — you need a store every teammate's Claude can reach. Connect your assistants, and let the first decision your Claude writes down be the one your teammate's Claude reads tomorrow.