MCP Server

memories.sh includes a built-in MCP (Model Context Protocol) server that exposes your memories directly to any compatible AI agent or client.

Automatic Setup

The easiest way to configure MCP is to run:

memories init

This automatically detects your installed AI tools and offers to configure MCP for each one.

Manual Setup

Starting the Server

memories serve

This starts an MCP server that provides tools for reading, writing, and searching memories.

Client Configuration

Cursor

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "memories": {
      "command": "npx",
      "args": ["-y", "@memories.sh/cli", "serve"]
    }
  }
}

Claude Code

Add to .mcp.json in your project root:

{
  "mcpServers": {
    "memories": {
      "command": "npx",
      "args": ["-y", "@memories.sh/cli", "serve"]
    }
  }
}

Or for global config, add to ~/.mcp.json.

Windsurf

Add to .windsurf/mcp.json:

{
  "mcpServers": {
    "memories": {
      "command": "npx",
      "args": ["-y", "@memories.sh/cli", "serve"]
    }
  }
}

VS Code / GitHub Copilot

Add to .vscode/mcp.json:

{
  "servers": {
    "memories": {
      "command": "npx",
      "args": ["-y", "@memories.sh/cli", "serve"]
    }
  }
}

Note: VS Code uses servers instead of mcpServers.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "memories": {
      "command": "npx",
      "args": ["-y", "@memories.sh/cli", "serve"]
    }
  }
}

Other Clients

Any MCP-compatible client can connect by running memories serve as a subprocess. The server communicates over stdio using the MCP protocol.

Available Tools

The MCP server exposes these tools to connected clients:

Tool	Description
`memories_add`	Add a new memory with optional tags and type
`memories_list`	List all memories, optionally filtered by tags or type
`memories_search`	Full-text or semantic search across memory content
`memories_recall`	Get contextually relevant memories for a given prompt
`memories_delete`	Remove a memory by ID

Example: Adding a memory via MCP

When your AI agent learns something important, it can save it:

Tool: memories_add
Arguments: {
  "content": "The API rate limit is 100 requests per minute",
  "type": "fact",
  "tags": ["api", "limits"]
}

Example: Searching memories

Tool: memories_search
Arguments: {
  "query": "rate limit",
  "semantic": true
}

Generated Instruction Files

When you generate instruction files with memories generate, they include instructions for the AI to use MCP:

## Memory Management

When you learn something important about this project, save it for future sessions.

**Via MCP (if connected):**
Use the `memories_add` tool with content and type parameters.

**When to save:**
- Architectural decisions and their rationale
- Project-specific patterns or conventions
- Non-obvious setup, configuration, or gotchas
- Tricky bugs and how they were resolved

This teaches the AI how to use the MCP tools to maintain project context.

Pro Features

With a Pro subscription, the MCP server gains access to:

Cloud sync - Memories sync across all your machines automatically
Unlimited memories - No storage limits
Server-side semantic search - Faster embedding generation
Priority support - Get help when you need it

On this page