MCP Memory Server

A Model Context Protocol (MCP) server that provides semantic memory and codebase search using LanceDB and Voyage AI embeddings.

Features

- Semantic Search: Search your codebase and memories using natural language
- Auto-indexing: Automatically index your codebase files with embeddings
- Memory Storage: Store and retrieve important context across sessions
- Pattern Recognition: Extract recurring patterns from stored memories
- Project Context: Get structured context about your projects

Installation

$3

- Node.js >= 18.0.0
- A Voyage AI API key

$3

Add to your Claude Code configuration (~/.claude.json):

``json
{
"mcpServers": {
"memory": {
"command": "npx",
"args": ["-y", "@indokq/memory-mcp"],
"env": {
"VOYAGE_API_KEY": "your-voyage-api-key"
}
}
}
}
`

Tools

$3

Search past context and memories by semantic similarity. Also searches indexed codebase files.

Parameters:
- query (string, required): Search query text
- top_k (number, optional): Number of results to return (default: 5)
- threshold (number, optional): Minimum similarity threshold (default: 0.4)
- layers (string[], optional): Filter by memory layers (raw, consolidated, meta)
- project (string, optional): Filter by project name
- include_codebase (boolean, optional): Also search indexed codebase files (default: true)

$3

Store important information as a memory for future retrieval.

Parameters:
- content (string, required): Content to store as memory
- tags (string[], optional): Tags for categorization
- importance (number, optional): Importance score 0-1 (default: 0.5)
- project (string, optional): Project name
- layer (enum, optional): Memory layer - "raw", "consolidated", or "meta" (default: raw)

$3

Get recurring coding patterns and preferences from past sessions.

Parameters:
- project (string, optional): Filter patterns by project

$3

Get context about a specific project including structure, tech stack, and recent decisions.

Parameters:
- project (string, required): Project name to get context for

$3

Index codebase files for semantic search. Scans files and generates embeddings.

Parameters:
- path (string, optional): Root path to index (defaults to current directory)
- patterns (string[], optional): Glob patterns for files to include
- ignore (string[], optional): Glob patterns for files to ignore

Configuration

$3

- VOYAGE_API_KEY (required): Your Voyage AI API key

$3

All data is stored locally in ~/.claude/memory/lancedb/.

Models Used

- Embeddings: voyage-4-lite (1024 dimensions)
- Reranking: rerank-2.5-lite`

License

MIT