Set Up an AI Memory MCP Server for Claude Desktop

What is MCP?

The Model Context Protocol (MCP) is an open standard that lets AI applications like Claude Desktop, Cursor, and Windsurf connect to external tools and data sources. An MCP server provides tools that the AI can call during conversations.

Mengram's MCP server gives Claude Desktop 12 memory tools — search, add, profile, knowledge graph, triggers, and more — turning it into an AI that remembers everything across sessions.

Installation

You need a Mengram API key (get one free) and Claude Desktop installed.

Option 1: npx (recommended)

Add this to your Claude Desktop config file (claude_desktop_config.json):

{{
  "mcpServers": {{
    "mengram": {{
      "command": "npx",
      "args": ["-y", "mengram"],
      "env": {{
        "MENGRAM_API_KEY": "mg-your-api-key"
      }}
    }}
  }}
}}

Option 2: pip

pip install mengram-ai

{{
  "mcpServers": {{
    "mengram": {{
      "command": "python",
      "args": ["-m", "mengram", "mcp"],
      "env": {{
        "MENGRAM_API_KEY": "mg-your-api-key"
      }}
    }}
  }}
}}

Available tools (12 total)

Once connected, Claude Desktop gains these tools:

• memory_add — Store new memories from the conversation
• memory_search — Search across all memory types with semantic matching
• memory_profile — Generate a Cognitive Profile system prompt
• memory_list — List all memories for a user
• memory_delete — Remove specific memories
• memory_graph — Query the knowledge graph for entity relationships
• memory_triggers — Set up smart triggers that fire on memory events
• memory_import — Import from ChatGPT exports, Obsidian vaults, or text files
• memory_export — Export all memories as JSON
• memory_stats — View memory usage statistics
• memory_reflect — Trigger AI reflection on stored memories
• memory_deduplicate — Clean up duplicate or conflicting memories

How Claude uses memory

After setup, Claude Desktop automatically:

1. Searches your memory at the start of conversations for relevant context
2. Stores important information from your conversations
3. Uses your Cognitive Profile to personalize responses
4. Builds a knowledge graph of entities and relationships from your interactions

Example conversation

You: "Remember that I prefer using Railway for deployments and my project uses FastAPI"

Claude: I've stored that in your memory. Next time you ask about deployment,
I'll know you use Railway with FastAPI.

--- (next session) ---

You: "How should I set up CI/CD?"

Claude: Since you use Railway with FastAPI, here's how I'd set up your CI/CD...
[Uses memory context to give a personalized answer]

Also works with

The same MCP server works with Cursor, Windsurf, VS Code Copilot, and any other MCP-compatible client. The configuration is the same — just add the mengram server to your MCP config.

Also available as a Python/JS SDK for custom integrations.