A
securityA
licenseA
qualityA Model Context Protocol server that provides access to MongoDB databases. This server enables LLMs to inspect collection schemas and execute read-only queries.
Last updated -
8
280
263
MIT License
Provides persistent storage of conversation memories using MongoDB database with support for saving, retrieving, adding, and clearing memory records.
Memory MCP
A Model Context Protocol (MCP) server for logging and retrieving memories from LLM conversations with intelligent context window caching capabilities.
Features
Save Memories: Store memories from LLM conversations with timestamps and LLM identification
Retrieve Memories: Get all stored memories with detailed metadata
Add Memories: Append new memories without overwriting existing ones
Clear Memories: Remove all stored memories
Context Window Caching: Archive, retrieve, and summarize conversation context
Relevance Scoring: Automatically score archived content relevance to current context
Tag-based Search: Categorize and search context by tags
Conversation Orchestration: External system to manage context window caching
MongoDB Storage: Persistent storage using MongoDB database
Installation
Install dependencies:
npm install
Build the project:
npm run build
Configuration
Set the MongoDB connection string via environment variable:
export MONGODB_URI="mongodb://localhost:27017"
Default: mongodb://localhost:27017
Usage
Running the MCP Server
Start the MCP server:
npm start
Running the Conversation Orchestrator Demo
Try the interactive CLI demo:
npm run cli
The CLI demo allows you to:
Add messages to simulate conversation
See automatic archiving when context gets full
Trigger manual archiving and retrieval
Create summaries of archived content
Monitor conversation status and get recommendations
Basic Memory Tools
save-memories: Save all memories to the database, overwriting existing ones
memories: Array of memory strings to savellm: Name of the LLM (e.g., 'chatgpt', 'claude')userId: Optional user identifier
get-memories: Retrieve all memories from the database
No parameters required
add-memories: Add new memories to the database without overwriting existing ones
memories: Array of memory strings to addllm: Name of the LLM (e.g., 'chatgpt', 'claude')userId: Optional user identifier
clear-memories: Clear all memories from the database
No parameters required
Context Window Caching Tools
archive-context: Archive context messages for a conversation with tags and metadata
conversationId: Unique identifier for the conversationcontextMessages: Array of context messages to archivetags: Tags for categorizing the archived contentllm: Name of the LLM (e.g., 'chatgpt', 'claude')userId: Optional user identifier
retrieve-context: Retrieve relevant archived context for a conversation
conversationId: Unique identifier for the conversationtags: Optional tags to filter byminRelevanceScore: Minimum relevance score (0-1, default: 0.1)limit: Maximum number of items to return (default: 10)
score-relevance: Score the relevance of archived context against current conversation context
conversationId: Unique identifier for the conversationcurrentContext: Current conversation context to compare againstllm: Name of the LLM (e.g., 'chatgpt', 'claude')
create-summary: Create a summary of context items and link them to the summary
conversationId: Unique identifier for the conversationcontextItems: Context items to summarizesummaryText: Human-provided summary textllm: Name of the LLM (e.g., 'chatgpt', 'claude')userId: Optional user identifier
get-conversation-summaries: Get all summaries for a specific conversation
conversationId: Unique identifier for the conversation
search-context-by-tags: Search archived context and summaries by tags
tags: Tags to search for
Example Usage in LLM
Basic Memory Operations
Save all memories (overwrites existing):
User: "Save all my memories from this conversation to the MCP server" LLM: [Uses save-memories tool with current conversation memories]Retrieve all memories:
User: "Get all my memories from the MCP server" LLM: [Uses get-memories tool to retrieve stored memories]
Context Window Caching Workflow
Archive context when window gets full:
User: "The conversation is getting long, archive the early parts" LLM: [Uses archive-context tool to store old messages with tags]Score relevance of archived content:
User: "How relevant is the archived content to our current discussion?" LLM: [Uses score-relevance tool to evaluate archived content]Retrieve relevant archived context:
User: "Bring back the relevant archived information" LLM: [Uses retrieve-context tool to get relevant archived content]Create summaries for long conversations:
User: "Summarize the early parts of our conversation" LLM: [Uses create-summary tool to condense archived content]
Conversation Orchestration System
The ConversationOrchestrator class provides automatic context window management:
Key Features
Automatic Archiving: Archives content when context usage reaches 80%
Intelligent Retrieval: Retrieves relevant content when usage drops below 30%
Relevance Scoring: Uses keyword overlap to score archived content relevance
Smart Tagging: Automatically generates tags based on content keywords
Conversation State Management: Tracks active conversations and their context
Recommendations: Provides suggestions for optimal context management
Usage Example
import { ConversationOrchestrator } from "./orchestrator.js";
const orchestrator = new ConversationOrchestrator(8000); // 8k word limit
// Add a message (triggers automatic archiving/retrieval)
const result = await orchestrator.addMessage(
"conversation-123",
"This is a new message in the conversation",
"claude",
);
// Check if archiving is needed
if (result.archiveDecision?.shouldArchive) {
await orchestrator.executeArchive(result.archiveDecision, result.state);
}
// Check if retrieval is needed
if (result.retrievalDecision?.shouldRetrieve) {
await orchestrator.executeRetrieval(result.retrievalDecision, result.state);
}
Database Schema
Basic Memory Structure
type BasicMemory = {
_id: ObjectId;
memories: string[]; // Array of memory strings
timestamp: Date; // When memories were saved
llm: string; // LLM identifier (e.g., 'chatgpt', 'claude')
userId?: string; // Optional user identifier
};
Extended Memory Structure (Context Caching)
type ExtendedMemory = {
_id: ObjectId;
memories: string[]; // Array of memory strings
timestamp: Date; // When memories were saved
llm: string; // LLM identifier
userId?: string; // Optional user identifier
conversationId?: string; // Unique conversation identifier
contextType?: "active" | "archived" | "summary";
relevanceScore?: number; // 0-1 relevance score
tags?: string[]; // Categorization tags
parentContextId?: ObjectId; // Reference to original content for summaries
messageIndex?: number; // Order within conversation
wordCount?: number; // Size tracking
summaryText?: string; // Condensed version
};
Context Window Caching Workflow
The orchestration system automatically:
Monitors conversation length and context usage
Archives content when context usage reaches 80%
Scores relevance of archived content against current context
Retrieves relevant content when usage drops below 30%
Creates summaries to condense very long conversations
Key Features
Conversation Grouping: All archived content is linked to specific conversation IDs
Relevance Scoring: Simple keyword overlap scoring (can be enhanced with semantic similarity)
Tag-based Organization: Categorize content for easy retrieval
Summary Linking: Preserve links between summaries and original content
Backward Compatibility: All existing memory functions work unchanged
Automatic Management: No manual intervention required for basic operations
Development
To run in development mode:
npm run build
node build/index.js
To run the CLI demo:
npm run cli
License
ISC
This server cannot be installed
hybrid server
The server is able to function both locally and remotely, depending on the configuration or use case.
A Model Context Protocol server that allows saving, retrieving, adding, and clearing memories from LLM conversations with MongoDB persistence.
Related MCP Servers
- -securityAlicense-qualityA Model Context Protocol server that enables LLMs to interact directly with MongoDB databases, allowing users to query collections, inspect schemas, and manage data through natural language.Last updated -1001MIT License
- -securityAlicense-qualityA Model Context Protocol server that enables LLMs to interact directly with MongoDB databases, allowing users to query collections, inspect schemas, and manage data through natural language.Last updated -100MIT License
- -securityAlicense-qualityA Model Context Protocol (MCP) server that enables LLMs to interact directly with MongoDB databases, allowing them to query collections, inspect schemas, and manage data seamlessly through natural language.