Unified Memory Adapter

Unified Memory Adapter is an MCP server that provides AI assistants with persistent memory capabilities, enabling them to store, retrieve, and manage contextual information across conversations and sessions. This MCP bridge allows language models to maintain knowledge graphs, user preferences, project context, and conversation history that persists beyond individual chat sessions, creating a durable memory layer that enhances AI assistant continuity and personalization.

When to Use This MCP Server

Connect this server when...

• You want your AI assistant to remember context, preferences, and decisions across multiple conversation sessions
• Your workflow involves long-running projects where the AI needs to recall previous discussions and conclusions
• You are building AI-powered workflows that require persistent state management beyond single interactions
• You need the AI to maintain a knowledge graph of entities, relationships, and facts relevant to your work
• You want to share contextual knowledge between different AI assistant sessions or team members

Consider alternatives when...

• Your interactions are single-shot queries that do not benefit from cross-session memory
• You need structured database storage rather than AI-oriented knowledge management
• Your memory requirements are met by in-context conversation history without persistence

Quick Start

Copy
# .mcp.json configuration
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "env": {
        "MEMORY_STORAGE_PATH": "~/.ai-memory/knowledge.json"
      }
    }
  }
}

Connection setup:

1. Ensure Node.js 18+ is installed on your system
2. Choose a storage location for your memory file (defaults to the home directory)
3. Add the configuration above to your .mcp.json file
4. Restart your MCP client to activate the memory server

Example tool usage:

# Store a fact
> Remember that the production database is PostgreSQL 15 running on AWS RDS in us-east-1

# Retrieve knowledge
> What do you know about our production database setup?

# Create entity relationships
> Note that the user-service depends on the auth-service and the database-service

Core Concepts

Concept	Purpose	Details
Knowledge Graph	Structured memory	A graph of entities and relationships that represents stored knowledge with typed connections
Entities	Knowledge nodes	Named objects (people, services, concepts) with properties and descriptions stored in the graph
Relations	Entity connections	Typed relationships between entities that capture how things connect and depend on each other
Observations	Factual records	Individual facts, preferences, or decisions attached to entities as timestamped observations
Persistence	Cross-session storage	Knowledge graph is serialized to disk and loaded on startup, surviving between sessions

Architecture:

+------------------+       +------------------+       +------------------+
|  Local Storage   |       |  Memory MCP      |       |  AI Assistant    |
|  (JSON file)     |<----->|  Server (npx)    |<----->|  (Claude, etc.)  |
|  ~/.ai-memory/   | fs    |  stdio transport  | stdio |                  |
+------------------+       +------------------+       +------------------+
        |
        v
+------------------------------------------------------+
|  Entities > Relations > Observations > Queries        |
+------------------------------------------------------+

Configuration

Parameter	Type	Default	Description
MEMORY_STORAGE_PATH	string	~/.ai-memory/knowledge.json	File path where the knowledge graph is persisted between sessions
auto_save	boolean	true	Automatically save the knowledge graph after each modification
max_entities	integer	10000	Maximum number of entities the knowledge graph can hold
backup_count	integer	3	Number of backup copies maintained for the knowledge graph file
search_depth	integer	3	Maximum traversal depth for relationship queries in the knowledge graph

Best Practices

1. Establish naming conventions for entities. Use consistent naming patterns for entities (kebab-case for services, PascalCase for concepts) so the AI can reliably find and reference stored knowledge. Inconsistent naming leads to duplicate entities that fragment your knowledge base.

2. Periodically review and prune stored knowledge. Over time, the knowledge graph can accumulate outdated or incorrect information. Schedule periodic reviews to remove stale entities, update changed facts, and consolidate duplicate entries that may have been created across sessions.

3. Use typed relationships for structured queries. When storing relationships between entities, use descriptive relationship types like "depends-on", "owned-by", "deployed-to" rather than generic "related-to" connections. Specific types enable more precise queries and richer knowledge graph traversal.

4. Separate project-specific and global knowledge. If you work on multiple projects, consider using separate memory storage paths for each project. This prevents knowledge from one project contaminating another and keeps each knowledge graph focused and manageable.

5. Back up your knowledge graph regularly. The memory file contains valuable accumulated knowledge. Ensure the backup_count is set appropriately and consider including the knowledge graph file in your backup strategy. Loss of the memory file means losing all stored context.

Common Issues

Knowledge graph file grows too large over time. As entities and observations accumulate, the JSON file can become large and slow to load. Prune unused entities, remove outdated observations, and consider archiving old knowledge. Set the max_entities limit to prevent unbounded growth.

AI creates duplicate entities with slightly different names. Without strict naming conventions, the AI may create "user-service", "UserService", and "user service" as separate entities. Establish and communicate naming rules, and periodically merge duplicate entities to maintain graph integrity.

Memory not persisting between sessions. Verify the MEMORY_STORAGE_PATH points to a writable directory. Check file permissions and ensure the directory exists. If auto_save is disabled, the knowledge graph is only saved when explicitly requested, which may lead to data loss on unexpected termination.