OPEN LINK ↗
// 66d agoINFRASTRUCTURE
MemoryGraph MCP Leaves Memories Hard to Find
MemoryGraph is a local-first MCP server that stores AI assistant memories in a searchable knowledge graph. This Reddit thread spotlights rough edges in visibility and setup, especially in LM Studio on Windows: get_memory works by ID while search_memory, recall_memory, and recent-activity counts still come back empty.
// ANALYSIS
- –Local-first memory only works if the surface area is trustworthy. MemoryGraph looks promising under the hood, but this thread shows how easily setup friction and weak observability can make a healthy backend feel broken.
- –The docs say MemoryGraph defaults to SQLite and keeps data local unless you opt into cloud or another graph backend, so an empty activity view is more likely a retrieval/indexing issue than missing storage.
- –Windows users are especially exposed to launcher friction; the memorygraph.exe workaround shows how much the client path and binary resolution matter.
- –If get_memory works while search_memory and recall_memory fail, the memory index is the problem the user actually experiences, even if the raw data is present.
- –Broad MCP support is a strength, but LM Studio, Claude, Cursor, and VS Code each add their own setup and debugging surface.
// TAGS
memorygraphmcpsearchagentself-hostedopen-source
DISCOVERED
66d ago
2026-03-22
PUBLISHED
66d ago
2026-03-22
RELEVANCE
8/ 10
AUTHOR
Kamushii--