@smara/mcp-server

MCP server for the Smara Memory API — give any AI app persistent, decay-aware memory.

Quick Start

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "smara": {
      "command": "npx",
      "args": ["-y", "@smara/mcp-server"],
      "env": {
        "SMARA_API_KEY": "your-api-key-here"
      }
    }
  }
}

Get a free API key at smara.io (10,000 memories, no credit card).

Tools

Tool	Description
`store_memory`	Store a fact about a user with importance scoring
`search_memories`	Semantic search with Ebbinghaus decay-aware ranking
`get_user_context`	Pre-formatted context string for LLM system prompts
`delete_memory`	Delete a specific memory
`get_usage`	Check plan limits and memory count

How It Works

Smara combines vector similarity search (Voyage AI embeddings) with Ebbinghaus forgetting curves. Memories decay over time — recent, frequently-accessed memories rank higher, just like human recall.

score = similarity × 0.7 + decay_score × 0.3

Works With

Claude Desktop
Claude Code
Cursor
Windsurf
VS Code (GitHub Copilot)
Any MCP-compatible client

Environment Variables

Variable	Required	Default
`SMARA_API_KEY`	Yes	—
`SMARA_API_URL`	No	`https://api.smara.io`

License

MIT

Smara Memory MCP Server