Neural memory MCP Server

🧠 Neural Memory v2.0

⚠️ IMPORTANT NOTICE: This is an experimental MCP (Model Context Protocol) tool. While it has shown promising results when used locally, it is still under active development and may evolve over time. Feedback and contributions are highly appreciated! Please report any issues, suggestions, or feature requests on our GitHub Issues page. Your input helps make this tool better for everyone! nya~

---

Model Context Protocol (MCP) system for knowledge memorization and contextual retrieval.

---

🚀 Try Neural Memory NOW!

Want to try it right away in your Cline? Add this configuration to your cline_mcp_settings.json:

{
  "mcpServers": {
    "neural-memory": {
      "timeout": 120,
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "@drakonkat/neural-memory@latest"
      ]
    }
  }
}

For Local Development

{
  "mcpServers": {
    "neural-memory": {
      "timeout": 120,
      "type": "stdio",
      "command": "node",
      "args": [
        "./neural-memory/src/index.js"
      ]
    }
  }
}

⚠️ Note: Replace the path with your local project path!

---

What's New in v2.0

🎯 Unified Database

• No more separate databases per project! All memory resides in a single SQLite database
• Faster, simpler, zero configuration

🎓 Skills Framework

• Rigid schema for registering skills with consistent structure
• framework, language, filePattern, learnSteps, useCases
• Automatic context-based suggestions

📊 HTML Reports

• Generate visual reports of your memory
• Statistics, most used skills, recent work

Quick Start

import { initialize, handleMcpRequest } from 'neural-memory';

// Initialize the service
await initialize();

// Add a node
await handleMcpRequest('add_node', {
  keywords: ['fastify', 'middleware', 'auth'],
  content: 'Implementation of JWT middleware...',
  type: 'generic'
});

// Register a skill
await handleMcpRequest('register_skill', {
  name: 'Fastify CRUD API',
  framework: 'fastify',
  language: 'javascript',
  filePattern: '**/*.service.js',
  learnSteps: [
    '1. Initialize project',
    '2. Create validation schema',
    '3. Implement handler'
  ],
  useCases: [
    'Create REST API',
    'Handle standardized errors'
  ]
});

// Search for skills
const skills = await handleMcpRequest('suggest_skills', {
  currentKeywords: ['fastify', 'api'],
  domain: 'javascript'
});

// Save context
await handleMcpRequest('save_context_snapshot', {
  summary: 'Refactoring 80% complete',
  pendingTasks: ['Integration testing', 'Deploy to staging'],
  learnings: ['Prisma requires explicit migrations']
});

// Generate report
const report = await handleMcpRequest('get_memory_report', {
  format: 'html'
});

Available Tools

Node Management

Tool	Description
add_node	Add node to memory
search_nodes	Search nodes (with FTS5 + confidence)
delete_node	Delete node

Skills Framework

Tool	Description
register_skill	Register skill (rigid schema)
suggest_skills	Suggest skills

Context Management

Tool	Description
save_context_snapshot	Save context snapshot
restore_context	Restore context

Reports

Tool	Description
get_memory_report	JSON/HTML report

Node Types

Type	Priority	Use Case
skill	⭐⭐⭐	Knowledge skill
error	⭐⭐⭐	Resolved error
operation	⭐⭐	How-to
convention	⭐⭐	Naming rules
edge_case	⭐⭐	Edge case
pattern	⭐	Design pattern
task	⭐	Completed task
generic	⭐	Generic node

Confidence Scoring

The system calculates a confidence score (0.0-1.0) for each result based on:

• BM25 Score (35%) - Full-text search ranking
• Keyword Match (25%) - Keywords overlap
• Recency Bonus (10%) - Recent nodes preferred
• Type Score (15%) - Skills/errors are more important
• Weight (15%) - Manual weight

Rigid Skill Schema

{
  "name": "Skill Name",
  "framework": "fastify|react|prisma|...",
  "language": "javascript|typescript|python|...",
  "filePattern": "**/*.service.js",
  "learnSteps": ["Step 1", "Step 2"],
  "useCases": ["Case 1", "Case 2"],
  "implementation": "// Code...",
  "examples": ["file1.js"],
  "prerequisites": ["Prerequisite 1"]
}

Installation

npm install neural-memory

Database

The unified database is located at:

data/neural-memory-unified.sqlite

License

MIT