icon for mcp server

Like-I-Said

STDIO

Persistent memory server for AI assistants to remember conversations across sessions

Like-I-Said MCP v2

Like-I-Said v2: Task Management & Memory for Claude Desktop

npm version License: MIT

Turn Claude Desktop into your intelligent project manager. Track tasks, remember context, and maintain continuity across sessions.

🚀 v3.0.0 Released - Socket Architecture & Intelligence Upgrade!

🔥 Major Release (August 2025): Version 3.0.0 solves duplicate process issues and adds AI features:

  • 🆕 Socket-Based Architecture - HTTP/SSE mode prevents duplicate processes (fixes API Error 500!)
  • ✅ Fuzzy Search - Typo-tolerant search with Fuse.js (handles "serch" → "search")
  • ✅ Universal Work Detector - Automatically captures work patterns and creates memories
  • ✅ Enhanced Search - Multi-mode search combining exact, expanded, semantic, and fuzzy results
  • ✅ Windows Integration - Improved cross-platform support with Windows-specific scripts

🔒 No More Duplicates: Port binding ensures only ONE instance can run! 📊 Search Intelligence: Now finds 700+ memories with typo-tolerant matching! 🤖 Work Detection: Automatically identifies problem-solving, implementation, and configuration work!

# New HTTP mode (prevents duplicates!)
export MCP_TRANSPORT=http
node server-markdown.js

# Or classic stdio mode
npm install @endlessblink/[email protected]

📚 Table of Contents

🎯 What Makes This Different

Like-I-Said v3.0 brings AI-powered intelligence to memory and task management:

🧠 Intelligent Search & Discovery

  • Fuzzy Search - Finds memories even with typos ("serch" finds "search")
  • Multi-Mode Search - Combines exact, expanded, semantic, and fuzzy results
  • Smart Relevance - Prioritizes results based on context and usage patterns
  • Typo Tolerance - Never lose memories due to spelling mistakes

🤖 Universal Work Detector

  • Automatic Pattern Recognition - Identifies problem-solving, implementation, research activities
  • Smart Memory Creation - Captures important work without manual effort
  • Activity Monitoring - Tracks your workflow patterns intelligently
  • Safe Mode - Gradual integration that learns your preferences

📋 Advanced Task Management

  • Smart Task Tracking - Say "I finished the auth system" and Claude updates your tasks
  • Automatic Memory-Task Linking - Every decision, code snippet, and conversation links to relevant tasks
  • Natural Language Updates - No commands needed, just talk naturally about your progress
  • Cross-Session Continuity - Pick up exactly where you left off, even days later
  • Project-Based Organization - Keep multiple projects separate and organized

🚀 Installation

Prerequisites

Before installing Like-I-Said v2, you need to have Node.js installed on your system:

  • Node.js (version 14 or higher) - Download from nodejs.org
    • This includes NPM and NPX which are required for installation
    • To check if you have Node.js installed, run: node --version
    • If you see a version number (e.g., v18.17.0), you're ready to proceed
    • If not, download and install Node.js from the link above

Note: NPX commands require Node.js to be installed. If you see an error like "npx: command not found", please install Node.js first from https://nodejs.org/

Installation Methods

Option 1: Automatic Installation (Recommended)

Works for both Claude Desktop and Claude Code - one command does everything:

npx @endlessblink/like-i-said-v2@latest like-i-said-v2 install

# Or install to a specific directory:
npx @endlessblink/like-i-said-v2@latest like-i-said-v2 install --path /custom/path

This single command automatically:

  • Installs the MCP server completely (no additional commands needed)
  • Configures your Claude client (Desktop or Code)
  • Sets up necessary directories
  • No manual configuration or menu interaction required

Option 2: Claude Code Direct Registration

If you're using Claude Code and Option 1 didn't work:

claude mcp add like-i-said-memory-v2 -- npx -p @endlessblink/like-i-said-v2@latest like-i-said-v2

This registers the MCP server directly with Claude Code's configuration system.

Verify Installation

After installation:

  1. Restart your Claude client
  2. Ask Claude: "What MCP tools do you have available?"
  3. You should see 27 tools including add_memory, create_task, etc.

Example usage:

"Create a task: Build authentication system"
"Remember: We're using JWT tokens with refresh rotation"
"What am I currently working on?"

🖥️ Manual IDE Configuration

If automatic configuration fails, you can manually configure your IDE:

For Cursor

Option 1: Quick Setup (No Local Files)

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "like-i-said-memory-v2": {
      "command": "npx",
      "args": ["-y", "-p", "@endlessblink/like-i-said-v2@latest", "like-i-said-v2"]
    }
  }
}

Option 2: Local Installation

  1. First run: npx -p @endlessblink/like-i-said-v2@latest like-i-said-v2 install

  2. The installer will automatically configure Cursor for you

  3. Restart Cursor and verify all 27 tools are available

For Windsurf

Option 1: Quick Setup (No Local Files)

// ~/.codeium/windsurf/mcp_config.json
{
  "mcp": {
    "servers": {
      "like-i-said-memory-v2": {
        "command": "npx",
        "args": ["-y", "-p", "@endlessblink/like-i-said-v2@latest", "like-i-said-v2"]
      }
    }
  }
}

Option 2: Local Installation

  1. First run: npx -p @endlessblink/like-i-said-v2@latest like-i-said-v2 install

  2. The installer will automatically configure Windsurf for you

  3. Restart Windsurf and verify all 27 tools are available

For VS Code with Continue

Follow the Continue extension's MCP configuration guide to add the server.

📊 Running the Dashboard

The dashboard provides a visual interface for managing memories and tasks.

Starting the Dashboard

Quick Start (Recommended)

# Clone the repository
git clone https://github.com/endlessblink/Like-I-Said-memory-mcp-server.git
cd Like-I-Said-memory-mcp-server

# Windows
scripts\run-dashboard.bat

# Mac/Linux
./scripts/run-dashboard.sh

Manual Start

# If you already have the repo cloned
cd Like-I-Said-memory-mcp-server
npm install
npm run start:dashboard

Accessing the Dashboard

⚡ IMPORTANT: The dashboard URL is shown when you start the server!

  1. Look for "DASHBOARD READY! Access it at:" in the console
  2. Open the URL shown (usually http://localhost:3001)
  3. IGNORE port 5173 - that's only for development
  4. The dashboard shows all memories and tasks with real-time updates
  5. Features include:
    • 📊 Task progress visualization
    • 🔍 Search and filter memories
    • 🔗 View task-memory relationships
    • 📁 Manage projects
    • 📤 Export/import data
    • 📈 Analytics and insights

💪 Key Features

Task Management That Understands You

  • Status tracking - todo → in_progress → done → blocked
  • Smart updates - "I'm done with X" automatically updates status
  • Subtask management - Break down complex tasks
  • Progress analytics - See what you've accomplished

Memory With Purpose

  • Task-linked memories - Every memory connects to relevant tasks
  • Project organization - Separate contexts for different projects
  • Semantic search - Find information using natural language
  • Automatic categorization - Code, decisions, research, etc.

Built for Real Work

  • Dynamic path configuration - "Change my task folder to D:\Projects"
  • Environment variables - Set MEMORY_DIR and TASK_DIR
  • Windows optimized - Full Windows path support
  • Real-time dashboard - Visual interface (see console for URL)

🛠️ Available Tools (27 Total)

Core Task Tools

  • create_task - Create tasks with automatic memory linking
  • update_task - Update status, add subtasks, link memories
  • smart_status_update - Natural language status changes ("I finished X")
  • list_tasks - View tasks by status, project, or priority
  • get_task_context - See all memories and subtasks for a task
  • get_task_status_analytics - Productivity insights and suggestions

Memory Tools

  • add_memory - Store any information with rich metadata
  • search_memories - Find information across all projects
  • list_memories - Browse memories by project or category
  • generate_dropoff - Create handoff documents for new sessions

Path Management (New in v2.6+)

  • set_memory_path - Claude can change where memories are stored
  • set_task_path - Claude can change where tasks are stored
  • get_current_paths - Check current storage locations

AI Enhancement Tools

  • enhance_memory_metadata - Generate titles and summaries
  • batch_enhance_memories - Bulk enhance multiple memories
  • validate_task_workflow - Check workflow validity
  • get_automation_suggestions - Get automation ideas
  • Plus 7 more enhancement tools...

📁 How It Works

Tasks and memories are stored as markdown files with rich metadata:

memories/
├── project-name/
│   ├── 2024-07-15-auth-decision-a7b3c9.md
│   └── 2024-07-15-jwt-implementation-d4e8f2.md
tasks/
├── project-name/
│   ├── TASK-00001-build-authentication.md
│   └── TASK-00002-add-user-dashboard.md

Each task tracks:

  • Status, priority, and timestamps
  • Linked memories and subtasks
  • Project and category assignment
  • Natural language descriptions

🎮 Advanced Usage

Natural Language Task Management

"I'm blocked on the payment integration because we need API credentials"
"The frontend is done but needs code review"
"What should I work on next?"
"Show me my progress this week"

Dynamic Path Configuration

"Change my memory storage to D:\AI\Memories"
"Set task directory to C:\Projects\MyApp\tasks"
"Where are my files currently stored?"

Project Switching

"Switch to the mobile-app project"
"Show me all tasks for the API project"
"What's the status across all my projects?"

🐛 Troubleshooting

"Server not connecting"

# Update to latest version
npx -p @endlessblink/like-i-said-v2@latest like-i-said-v2 install

# For Claude Code users (no local files)
claude mcp add like-i-said-memory-v2 -- npx -p @endlessblink/like-i-said-v2@latest like-i-said-v2

"Can't find memories after path change"

Your old memories stay in the original location. Either:

  • Copy them to the new location
  • Change back to the original path

Debug Mode

Set environment variable before starting Claude:

set DEBUG_MCP=true  # Windows
export DEBUG_MCP=true  # Mac/Linux

Common Issues

Windows Path Issues:

  • Use forward slashes: C:/Users/name/memories
  • Or escape backslashes: C:\\Users\\name\\memories

Port Conflicts:

  • Dashboard API uses port 3001
  • Development UI uses port 5173
  • Change with PORT environment variable if needed

🤝 Contributing

Found a bug? Have a feature idea?

📄 License

MIT © endlessblink

Note: This is an MCP (Model Context Protocol) server for Claude Desktop. It requires Claude Desktop or a compatible MCP client to function.

Be the First to Experience MCP Now