The memory-mcp platform is currently in development. The Self-hosted Community Edition is available now!

get_memory

Direct retrieval by ID — no search, no scoring. Returns the complete memory record including the bidirectional related memory graph built at store time.

Parameters #

ParameterTypeRequiredDescription
memory_idnumberYesThe ID of the memory to retrieve

Response #

Beyond content, labels, and source, get_memory returns:

FieldDescription
related_memoriesMemories with similar content, pre-computed at store time — each with id, similarity, and source
meta.timestampExact storage timestamp
meta.embedding_tablesWhich embedding model was used
timeHuman-readable age (“10 minutes ago”, “4 days ago”)

When a memory is stored, Memory MCP automatically computes similarity against existing memories and builds a bidirectional map of related entries. If memory #1008 is related to #478, both records carry that link — no orphaned references.

The graph self-maintains:

  • On store — new relationships computed and linked
  • On delete — back-references cleaned up automatically

This means get_memory gives you not just the memory you asked for, but a map of where it sits in the broader knowledge graph.


Examples #

# Retrieve a specific memory
get_memory(memory_id: 42)

# Follow a reference surfaced by retrieve_memories
retrieve_memories(query: "authentication decision")
→ Memory #42 returned with related_memories: [{id: 38}, {id: 51}]
→ get_memory(memory_id: 38)
→ get_memory(memory_id: 51)

Power Combinations #

Breadcrumb traversal — follow the related memory graph without semantic search. Zero embedding cost, pure graph navigation:

get_memory(memory_id: 478)
→ related_memories: [{id: 312, similarity: "83%"}, {id: 156, similarity: "75%"}]
→ get_memory(memory_id: 312)   # follow the strongest link
→ get_memory(memory_id: 156)   # explore the second branch

Useful when you know roughly where you are in the knowledge graph and want to explore the neighbourhood — past decisions, related discussions, follow-up exchanges — without constructing a query.

Context reconstruction — when an agent surfaces a memory ID (from recalled-memories, a previous response, or a note), get_memory retrieves the full record including surrounding context via related_memories. One ID becomes a web.

?

AI Assistant

0/500