Mirror Memory MCP Server

Give your AI a memory. Mirror Memory is a personal knowledge base that lets Claude, ChatGPT, Cursor, and any MCP-compatible AI remember what matters to you — your preferences, experiences, goals, people, and ideas.

Capture thoughts via the web app, Telegram, or directly through MCP tools. Search them semantically. Your AI assistant becomes one that actually knows you.

Website: mirrormemory.ai App: app.mirrormemory.ai

MCP Tools

Tool	Description
`search_memories`	Semantic similarity search across your memories. Use specific terms from the user's question for best results.
`store_memory`	Save a new thought or piece of information. Auto-categorized via AI, auto-extracts people, topics, action items, and dates.
`list_categories`	List all available memory categories (Ideas, Tasks, Notes, People, Work, etc.).
`get_recent_memories`	Retrieve your most recently stored memories.

Quick Setup

1. Get your API key

2. Connect your AI

Claude Desktop (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "mirror-memory": {
      "url": "https://hcelnlqmgddxzvjegpdr.supabase.co/functions/v1/mcp-server",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Claude Code (~/.claude/settings.json):

{
  "mcpServers": {
    "mirror-memory": {
      "url": "https://hcelnlqmgddxzvjegpdr.supabase.co/functions/v1/mcp-server",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "mirror-memory": {
      "url": "https://hcelnlqmgddxzvjegpdr.supabase.co/functions/v1/mcp-server",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Windsurf, Cline, or any MCP client — use the same URL and API key pattern above.

3. Start talking

Once connected, your AI will automatically search your memories for relevant context and store new information you share. No special commands needed.

How It Works

You say something → AI calls store_memory → embedded + categorized + metadata extracted → stored
You ask something → AI calls search_memories → semantic vector search → relevant memories returned

Every memory is:

Embedded as a 1536-dim vector for semantic search
Auto-categorized into one of 12+ categories via AI
Metadata-enriched — people mentioned, action items, topics, dates, and knowledge type are extracted automatically
Deduplicated via SHA-256 content fingerprinting

Features

Multi-channel capture — Web app, Telegram bot, MCP tools, import
Semantic search — Find memories by meaning, not just keywords
AI categorization — 12 built-in categories + custom categories
Rich metadata — People, action items, topics, dates extracted automatically
Memory lifecycle — Active, archived, and superseded states
Weekly digests — AI-synthesized email summaries of your memories
Export — Download everything as markdown. No lock-in.
Privacy first — Row-level security. We track engagement events, never content.

Pricing

Plan	Price	Memories	Channels
Free	$0	100/month	Web + Telegram
Solo Pro	$9/mo	Unlimited	Web + Telegram + MCP + Export
Lifetime	$199 once	Unlimited	Everything, forever

MCP access requires Solo Pro or Lifetime.

Transport

Protocol: MCP over Streamable HTTP (JSON-RPC 2.0)
Auth: API key in Authorization: Bearer <key> header
Rate limit: 100 requests/minute per key
Endpoint: https://hcelnlqmgddxzvjegpdr.supabase.co/functions/v1/mcp-server

License

Mirror Memory is a managed service. This repository contains documentation and configuration for the MCP server integration.