Neo4j Memory Server

MCPOpen SourceMIT45.9

by sylweriusz • Search & Retrieval

An MCP server that provides persistent, graph-based memory for AI assistants using Neo4j with semantic vector search and graph operations.

Example Use Cases

1
Persist and retrieve contextual memories as graph nodes with observations and metadata across project-specific Neo4j databases.
2
Semantic vector search combined with exact matching and graph traversal (via Neo4j GDS) to surface the most relevant memories.
3
Create and manage typed relationships, perform batch memory operations, and run temporal queries to track how knowledge evolves.

Description

This server stores memories as interconnected graph nodes and observations, enabling agents to remember, search, connect, and organize knowledge over time. It combines semantic vector search (via Neo4j GDS), exact matching, wildcards, and graph traversal in a unified API with batch operations and temporal filtering. Built as an MCP-native toolset, it integrates with Claude Desktop and supports multi-database project isolation. Note: the GDS plugin (e.g., DozerDB setup) is required for vector operations.

Capabilities(13 total)

Tools (4)

memory_store

Create memories with observations and relationships. **Pattern**: Search→Create→Connect. **Observations**: Self-contained context units (what/when/where/why/impact). **LocalIds**: Cross-references within THIS request only. **Limits**: 50 memories, 200 relations. **Quality**: Each observation = complete detective notes answering setting/action/actors/evidence/impact/significance.

memories:array*relations:arrayoptions:object

memory_find

Unified search/retrieval. **Query**: text, IDs array, or "*". **Context**: minimal (lists), full (everything), relations-only (graph). **Temporal**: createdAfter "7d"/"2024-01-15". **Graph**: traverseFrom + relations + depth. **Always search before creating**. Updates access timestamps for analytics.

query:any*limit:numbermemoryTypes:arrayincludeContext:stringthreshold:numberorderBy:stringcreatedAfter:stringcreatedBefore:stringmodifiedSince:stringaccessedSince:stringtraverseFrom:stringtraverseRelations:arraymaxDepth:numbertraverseDirection:string

memory_modify

Update/delete memories, manage observations/relations. **Operations**: update (properties), delete (cascade), add-observations (append insights), create-relations (link existing). **Atomic**: All succeed or all fail. **Quality**: One substantial observation per session - complete context stories, not fragments.

operation:string*target:stringtargets:arraychanges:objectobservations:arrayrelations:arrayoptions:object

database_switch

Switch active database context (creates if missing). ALL subsequent operations use this DB. Call once per session/project. Like 'cd' for memories. **Session-scoped**: Establishes context for entire workflow, not per-operation.

databaseName:string*

Prompts (9)

explore-memory-graph

starting_pointdepth

create-project-knowledge

project_namedomain

debug-orphaned-memories

memory_types

document-decision

decision_titlearea

analyze-memory-quality

scope

Quick Actions

View on GitHub

Security

Scanned 3 month(s) ago

Risk Level

MINIMAL

Read-only data retrieval, no side effects

Trust Score

C63/100

11/17 checks passed

Scores are informational only and provided “as is” without warranty. AgentHotspot assumes no liability for actions taken based on these ratings.

Quick Stats

Service TypeMCP

Pricing ModelFree

Capabilities4 Tools / 9 Prompts / 0 Resources

Ownersylweriusz

CategorySearch & Retrieval

DependenciesStandalone

Set Your Username