Like-I-Said-memory

Like-I-Said memory-mcp-server is a Node.js-based MCP that turns Claude Desktop into an integrated memory, task, and project-management assistant. It combines memory recall with task tracking, cross-session continuity, and intelligent search to help you manage projects and remember important decisions, code snippets, and conversations. The server exposes a suite of MCP tools (e.g., add_memory, create_task) that Claude can invoke to persist memories, link them to tasks, and surface context across sessions. With the v3.0 upgrade, you also gain a socket-based architecture and improved search capabilities, including typo-tolerant fuzzy search and multi-mode results, making it easier to locate relevant memories and tasks even if your queries aren’t exact. To use it, install the MCP server via npx as shown in the installation steps, then configure your Claude client to point at the server so Claude can access the 27 built-in tools for memory and task management.

Once running, you can interact with the server through Claude’s MCP integration, asking it to memory-tag decisions, attach notes to tasks, or pull up related memories when planning work. The dashboard provides a real-time view of memories and tasks, supports search and filters, and exposes task-memory relationships to help you understand the context behind decisions. The server also supports cross-session continuity so you can resume work where you left off, even after days apart, and it can automatically link decisions and code snippets to the appropriate tasks for easier traceability.

Prerequisites

• Node.js version 14 or higher (includes npm and npx). Verify with: node --version
• Internet access to install the MCP server via npm/pnpm/yarn

Installation steps

1. Install Node.js (if not already installed) from https://nodejs.org and verify: node --version npm --version

2. Install the MCP server using the recommended automatic installation:

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

This single command installs the MCP server, configures the Claude client, and sets up necessary directories.

3. If you prefer manual/more controlled setup, you can register the MCP server directly with Claude Code:

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

4. Start the dashboard (quick start):

# Clone and navigate to the repo, then start dashboard
git clone https://github.com/endlessblink/Like-I-Said-memory-mcp-server.git
cd Like-I-Said-memory-mcp-server
npm install
npm run start:dashboard

5. Verify installation by restarting Claude and asking: "What MCP tools do you have available?" You should see the 27 tools, including add_memory and create_task.

Tips and known considerations:

• The MCP server is Node.js-based and distributed via npm. The recommended run command uses npx with a specific package target to avoid duplicates.
• You can enable HTTP transport to prevent duplicate processes by setting MCP_TRANSPORT=http when starting the server. This helps avoid the common issue of multiple instances.
• The integrated dashboard will typically be available at http://localhost:3001 after startup; the development port 5173 is used only during development and should be ignored in production.
• If automatic configuration fails, you can manually wire in the server in your IDEs (Cursor, Windsurf, VS Code with Continue) using the provided JSON snippets in the README.
• The MCP tools expose cross-session memory-task linking, smart search, and automatic memory creation for work patterns detected by the Universal Work Detector feature introduced in the v3.0 release.