Learnlog MCP

🚀 learnlog-mcp

The MCP server that helps with metacognition by saving the prompts of LLM learning sessions locally.

Core Functionality

This project provides a tool to save LLM learning session prompts locally, enabling users to easily review their learning history, identify areas of confusion, and enhance metacognition.

🚀 Quick Start

Why was it created?

While studying with LLMs, I always had a lingering question:

"Do I really understand this?"

When reading AI explanations, it seems like I understand. I nod and think, "Ah, I see." But a few days later, when I try to explain the same concept to a teammate, I struggle to put it into words. Even though I've seen it many times, when it comes to speaking, all that remains is an abstract feeling.

That's when I realized that the feeling of understanding is different from true understanding.

It's easy to think we've "learned something" during a conversation with an LLM. However, when we look back at what we asked, we can see where we actually got stuck. The flow of questions reveals the gaps in our understanding.

So, I created this tool. A tool that allows me to easily save my questions and review them later. It's enough to collect just my short questions, rather than the long AI responses.

I plan to use this tool to explore DevOps and the open - source ecosystem in depth.

Why save prompts?

When learning with an LLM, we usually save the output (AI's response) in Notion or a notepad. But think about it:

• Given the same prompt, we can get a similar response anytime.
• Outputs are long and bulky, while prompts are short and light.
• What I asked is exactly what I don't know.

Traditional way: Session → Copy AI response → Save in Notion → Hard to find later

learnlog way: Session → Automatically save prompt → "What did I study today?" → Check immediately

Saving prompts helps us understand where we got stuck and what we were curious about. This is true metacognition.

Core Philosophy

1. Input > Output: Save prompts instead of outputs.
2. Lightweight: Store only light - weight questions instead of heavy responses.
3. Metacognition: Assess learning progress by asking "What did I ask?"
4. Session - independent: Consolidate learning records from any LLM and any session.

📦 Installation

1. Clone and build the repository

git clone https://github.com/YUJAEYUN/learnlog-mcp.git
cd learnlog-mcp
npm install
npm run build

2. Register with Claude Code

claude mcp add learnlog-mcp node /path/to/learnlog-mcp/dist/index.js

3. Register with Claude Desktop (Optional)

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

{
  "mcpServers": {
    "learnlog-mcp": {
      "command": "node",
      "args": ["/path/to/learnlog-mcp/dist/index.js"]
    }
  }
}

💻 Usage Examples

Automatic Saving

When you ask a learning - related question, the AI will automatically save the prompt. You don't need to say "Save it" separately.

Check Learning Records

"What did I study today?"
"What questions did I ask today?"

→ It will show all the prompts saved today in chronological order.

Query for a Specific Date

"What did I study on 2025-12-20?"

Check Recent Learning Status

"Show me the learning status of the last week"

→ It will summarize and show the number of questions asked each day.

Storage Location

~/learnlog/
└── prompts/
    ├── 2025-12-20.json
    ├── 2025-12-21.json
    └── 2025-12-22.json

Each file stores the prompts of the corresponding date in chronological order:

[
  {
    "prompt": "What is the N+1 query problem?",
    "timestamp": "2025-12-22T10:30:00.000Z"
  },
  {
    "prompt": "Then how do I implement eager loading?",
    "timestamp": "2025-12-22T10:35:00.000Z"
  }
]

Tool List

Usage Scenarios

1. Daily Learning Reflection

User: What did I study today?

AI: 📝 Today's questions (5)

1. [10:30] What is the N+1 query problem?
2. [10:35] How do I implement eager loading?
3. [14:20] When should I add an index?
4. [14:45] Why is the order of composite indexes important?
5. [16:00] How to interpret the EXPLAIN ANALYZE result?

→ I studied database optimization today!

2. Identify Learning Patterns

User: Show me the status of the last week

AI: 📊 Learning records in the last 7 days

• 2025-12-22: 5 questions
• 2025-12-21: 3 questions
• 2025-12-20: 8 questions

Total: 16 questions

3. Resume Previous Learning

User: What questions did I ask on December 20th?

AI: 📝 Questions on 2025-12-20 (8)

1. [09:00] What is the MCP protocol?
2. [09:15] What's the difference between Tools, Resources, and Prompts?
...

→ Oh, I studied MCP. I should continue asking questions!

📄 License

ISC