← Back to Home

Memori MCP Server

Security-first MCP server exposing Memori-backed remember and recall tools for IDE agents like Claude Code, Cursor, and Cline.

colygon/memori-mcp

Overview

memori-mcp is an MCP (Model Context Protocol) server that gives AI coding assistants persistent memory capabilities. It exposes two simple tools - remember and recall - backed by Memori's semantic memory layer.

The server is designed with security as the primary concern, following MCP security best practices to ensure safe operation in agentic environments.

Security Design

This server follows MCP security guidance and implements multiple layers of protection:

🎯 Least Privilege

Only two tools exposed. No file-system access, no network calls, no shell execution.

🔐 Scoped Memory

Memories are isolated by entity_id + process_id to prevent cross-project leakage.

📏 Input Limits

Size caps and validation on all tool inputs to prevent abuse and resource exhaustion.

🙈 No Secrets in Logs

Structured logging avoids dumping request payloads by default.

🚫 No Code Execution

Tools validate inputs and never execute arbitrary code (MCP-01 mitigation).

🔒 Local by Default

stdio transport with no open network listeners (MCP-05 mitigation).

Installation

From PyPI (Recommended)

Terminal

pip install memori-mcp

From Source

Terminal

git clone https://github.com/colygon/memori-mcp
cd memori-mcp
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
            

Dependencies

memori>=3.1.5 - Memori SDK for memory operations
mcp>=1.26.0 - Model Context Protocol SDK
pydantic>=2.5 - Input validation
sqlalchemy>=2.0 - Database abstraction

Configuration

Configure the server using environment variables:

Variable	Default	Description
`MEMORI_MCP_DB`	`sqlite:///./memori.db`	Database connection string
`MEMORI_MCP_ENTITY_ID`	Current OS user	User/entity identifier for memory scoping
`MEMORI_MCP_PROCESS_ID`	Current working directory name	Project/process identifier for memory scoping

Available Tools

memori.remember

Stores a memory snippet such as decisions, preferences, TODOs, or architectural notes.

# Example usage by the AI assistant
{
  "tool": "memori.remember",
  "arguments": {
    "content": "This project uses PostgreSQL with Prisma ORM",
    "type": "fact"  // preference | fact | summary | rule | todo
  }
}
                

memori.recall

Recalls relevant memories using semantic search based on the query.

# Example usage by the AI assistant
{
  "tool": "memori.recall",
  "arguments": {
    "query": "What database does this project use?",
    "limit": 5
  }
}
                

Client Configuration

Configure your AI coding assistant to use the Memori MCP server:

// ~/.claude/claude_desktop_config.json
{
  "mcpServers": {
    "memori": {
      "command": "memori-mcp",
      "env": {
        "MEMORI_MCP_DB": "sqlite:///~/.memori/claude.db",
        "MEMORI_MCP_ENTITY_ID": "${USER}"
      }
    }
  }
}
                

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "memori": {
      "command": "memori-mcp",
      "env": {
        "MEMORI_MCP_DB": "sqlite:///~/.memori/cursor.db"
      }
    }
  }
}
                

// VS Code settings.json
{
  "cline.mcpServers": {
    "memori": {
      "command": "memori-mcp",
      "env": {
        "MEMORI_MCP_DB": "sqlite:///~/.memori/cline.db"
      }
    }
  }
}
                

Running the Server

Start the server directly (uses stdio transport by default):

Terminal

memori-mcp

The server will start and listen for MCP messages on stdin/stdout. Your AI assistant client handles the connection automatically based on the configuration above.

Hardening Recommendations

For production deployments, consider these additional security measures:

Use dedicated databases per user/repo to prevent cross-project memory leakage
Run locally over stdio when possible (the default for Claude Code/Cursor/Cline)
If exposing over network, put it behind authentication and IP allowlists
Monitor logs for unusual tool usage patterns
Keep dependencies updated for security patches

Ready to Get Started?

Install the Memori MCP server and give your AI coding assistant persistent memory.

View on GitHub Integration Guide

Resources

GitHub Repository: github.com/colygon/memori-mcp
MCP Specification: modelcontextprotocol.io
MCP Security Guide: modelcontextprotocol-security.io
Memori Documentation: memorilabs.ai/docs
AI Assistants Guide: Integration walkthrough

Found a security issue? Please report it to security@memorilabs.ai

Back to Home