Proposal: optional read-only external memory MCP adapter

[LRSnowX]

Motivation

DevSpace is excellent at operating on the current workspace, but some users maintain long-running projects whose design decisions and handoff context live in external conversation-history systems. It would be useful to let DevSpace consume that context without coupling core workspace/domain code to any particular database or memory implementation.

Proposed shape

An opt-in, generic read-only memory adapter at the edge of the server:

• disabled by default
• configured with an external MCP command/backend
• registers only a small fixed read-only surface such as:
  • memory_search
  • memory_recent
  • memory_get_thread
  • memory_project_context
• every tool still requires workspaceId
• the adapter derives the project scope from the opened workspace instead of asking the model to guess it
• no arbitrary downstream tool forwarding, so a memory backend's write/admin tools remain unreachable through DevSpace
• implementation remains in a small adapter module; workspace/domain code stays unchanged

This is intentionally separate from subagents and does not require launching an agent/model to retrieve memory.

I have a tested v1.0.8 implementation in a fork (typecheck, full test suite, production build, and end-to-end MCP proxy test) and can prepare a small PR if this direction is welcome. I am not proposing to couple DevSpace specifically to one conversation-index product.