Claude Conversation Memory

A Model Context Protocol (MCP) server that gives Claude Code long-term memory by indexing your conversation history with semantic search, decision tracking, and mistake prevention.

💡 What It Does

• Remembers past conversations - Search your chat history with natural language
• Tracks decisions - Never forget why you made technical choices
• Prevents mistakes - Learn from past errors and avoid repeating them
• Links to git commits - Connect conversations to code changes
• Analyzes file history - See the complete evolution of files with context
• Migrates conversation history - Keep your history when renaming or moving projects
• Context transfer - Recall past work and apply it to current tasks (“remember X, now do Y based on that”)
• Forget selectively - Delete conversations by topic/keyword with automatic backups

⚠️ Important: Claude Code CLI Only

This MCP server works ONLY with Claude Code CLI.

It does NOT work with:

• ❌ Claude Desktop
• ❌ Claude Web
• ❌ Other Claude integrations

Claude Code CLI is required because it stores conversation history in ~/.claude/projects/ which this MCP indexes.

📦 Installation

Prerequisites

Required:

1. Claude Code CLI: https://github.com/anthropics/claude-code
2. Node.js: Version 18 or higher
3. sqlite-vec extension: Automatically loaded (bundled with the package)

Recommended for better semantic search quality: 4. Ollama: For high-quality local embeddings

# macOS/Linux
curl -fsSL https://ollama.com/install.sh | sh

# Or download from: https://ollama.com

5. Default embedding model (if using Ollama):

   # Pull the recommended model
   ollama pull mxbai-embed-large
   
   # Start Ollama service
   ollama serve
   

Note: Without Ollama, the MCP automatically falls back to Transformers.js (slower but works offline with no setup).

Install the MCP Server

npm install -g claude-conversation-memory-mcp

🎉 Automatic Configuration: The global installation will automatically configure the MCP server in Claude Code’s ~/.claude.json file. You’ll see a success message when it’s done!

Manual Configuration (if needed): If automatic configuration doesn’t work, see the Configure Claude Code CLI section below.

Discover Available Models: After installation, you can see all available embedding models and their dimensions:

• Run the CLI: claude-conversation-memory-mcp
• Type: config to see all available models organized by provider
• Or check the example config file: .claude-memory-config.example.jsonc

Configure Claude Code CLI

MCP Configuration File Priority:

Claude Code checks for MCP server configurations in this order (highest to lowest priority):

1. .mcp.json - Project-level (in your project root) - Highest Priority