Labsco
MCP SERVER

Work Memory MCP Server

by moontmsai

Store what you were working on with tags, a project and an importance score — then search it back, with version history and rollback.

Agent Memory
Summary
Versioned memory with a preview before you restore it.

Most memory servers let you overwrite and hope; here every change is recorded, restore can run in preview mode first, and a backup is taken before the rollback lands. The importance score is the other useful piece — searching with a minimum importance is a far better filter than recency when the store gets large.

What it is

A local work-memory store. Memories carry a project, tags, an importance score and a creator; search filters on all of them, and every change is versioned so an edit can be undone.

What you get
  • add_work_memory — store content with a project, tags, an importance score and a creator
  • search_work_memory — search with filters for project, tags, importance range, creator and date range, sorted by time, importance or relevance
  • list_work_memories — browse with filters and a choice of summary, detailed or minimal output
  • delete_work_memory — delete or archive a memory, with an explicit confirmation flag
  • get_related_keywords — keywords related to one you give, with a similarity threshold
  • get_search_stats — memory counts, popular tags, project distribution and index state
  • optimize_search_index — rebuild the index and clear orphaned entries
  • get_work_memory_history — change history filtered by memory, type (create, update, delete, archive), date, project or creator
  • get_work_memory_versions and list_memory_versions — the versions of a memory
  • restore_memory_version — roll back fully, selectively by field, or as a preview, with a backup taken first
  • get_server_status — server health, statistics and component checks
Requirements

Node.js 18 or higher. Install globally from npm as work-memory-mcp, clone and build, or take a release binary. Point the client at the built entry file and set WORK_MEMORY_DIR to the directory the data should live in; LOG_LEVEL, MAX_MEMORY_SIZE and the cache TTL are optional.

Setup effort

Build from source — clone the repository and build it, then point your client at the binary