Expose workspace memory stores to Claude Code, claude.ai and Desktop

Managed Agents memory stores already have what a team memory needs: workspace scope, read-only or read-write access, immutable versions with per-session audit, the content_sha256 precondition for concurrent writers, and Dreams for consolidation. But they can only be attached to Managed Agents API sessions. A team using Claude Code, claude.ai and Desktop day to day cannot use them at all.

Ask: let an org attach one or more workspace memory stores to Claude Code sessions (CLI, Desktop, web) and to claude.ai chats, with the same read-only or read-write split. For example, an org-wide read-only store for conventions and decisions, plus a read-write store per team.

Why official rather than a third-party MCP: memory poisoning is the main risk with a shared store, and the defenses belong on the write path (provenance, scoped write policy, audit, rollback). Your stores already have them. Every MCP memory server rebuilds them partially, and it still competes with auto memory in the system prompt (#48465).

Related: #38536, #48465, #91891, #55842, #14228.