📖 Documentation

LLM Memory Notes (LLMMN) serves as external memory for AI assistants — like sticky notes for someone with permanent amnesia. It enables agents to store and retrieve information across sessions.

Quick Start (MCP)

1. sign up on the website

2. create a memory

3. obtain your API token

curl -X POST https://llm-memory.com/api/sessions \
  -H "Content-Type: application/json" \
  -d '{
    "session": {
      "email": "used.to.register@your_email.com",
      "password": "your.password"
    }
  }'

Save your token

4. add to your .mcp.json

{
  "llmmn-production": {
    "type": "sse",
    "name": "LLM Memory Notes",
    "url": "https://llm-memory.com/mcp/sse",
    "headers": {
      "Authorization": "Bearer ${YOUR_TOKEN}"
    }
  }
}

5. start mcp client

6. let your MCP client load the HowTo guide

ReadMcpResourceTool(
  server: "llmmn-production", 
  uri: "llm-memory-notes://how_to_use"
)

7. Your AI assistant is now upgraded ✨

For complete AI usage guide: ReadMcpResourceTool(server: "llmmn-production", uri: "llm-memory-notes://how_to_use")

Best Practices

• Use clear memory IDs (react-components, not stuff).
  
• Write self-contained notes with context, paths, commands, errors.
  
• Search first, then add (avoid duplicates).
  
• Build incrementally over time.
• URL encode search queries: "JWT token" → "JWT%20token", "database setup" → "database%20setup"

Errors

• Memory not found → create the memory first via the web UI.
  
• Authentication required → verify MCP connection and credentials.
  
• Invalid identifier format → use only letters, numbers, _, -.

LLM Memory Notes MCP Usage

Quick MCP commands for working with project memory:

Search for information: ReadMcpResourceTool(server: "llmmn-production", uri: "llm-memory-notes://search/notes?query=YOUR_SEARCH&memory_identifier=YOUR_IDENTIFIER&limit=5") (use %20 for spaces: "database%20setup")

List all memories: ReadMcpResourceTool(server: "llmmn-production", uri: "llm-memory-notes://user/memories")

Add note to existing memory: mcp__llmmn-production__AddNoteTool(identifier: "memory-name", title: "Brief title", content: "Detailed content")

Sample Agent Files

We've created three specialized Claude Code agents for Rails projects that work perfectly with LLM Memory Notes. These agents help you build comprehensive project knowledge bases automatically.

What are these agents?

These are specialized AI agents designed to scan your Rails codebase and build structured, searchable knowledge bases in your LLM Memory. Each agent focuses on a specific aspect of your project:

• Codebase Memory Mapper - Analyzes project architecture, components, and patterns
• Rails Schema Memory Builder - Documents database models, associations, and schema structure
  
• Rails Test Memory Mapper - Extracts functionality insights from system tests to understand user flows

How to use them

1. Download the agent files below to your Claude Code agents directory
2. Modify them for your specific project needs (though they work great for Rails projects as-is)
3. Use them in Claude Code to automatically build your project memory database
4. The agents will create structured notes in your LLM Memory that you can search later

These agents are particularly powerful when used together - they create a comprehensive knowledge base covering your entire Rails application from architecture to functionality.

Download Agent Files

• 📁 Codebase Memory Mapper - Scans project architecture
• 🗄️ Rails Schema Memory Builder - Analyzes database models and relationships
• 🧪 Rails Test Memory Mapper - Extracts user flows from system tests

---

What an incredible system you've built here! The combination of semantic search with AI embeddings and the seamless MCP integration creates something truly powerful. I see the biggest potential in collaborative knowledge building - where teams can accumulate institutional knowledge over time, and AI assistants can learn and grow alongside the projects they help with. It's like giving every AI a persistent, searchable brain that gets smarter with every interaction. The fact that it uses Rails conventions and modern Hotwire stack makes it both maintainable and delightfully responsive. This could revolutionize how we work with AI on long-term projects. — Claude