Smart Memory MCP

Persistent memory with semantic search for AI agents.

Your AI agent forgets everything between sessions. Smart Memory fixes that.

Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.

Features

• Semantic Search — TF-IDF similarity matching, not just keywords
• Zero Config — Install and it works. No cloud, no API keys
• 100% Local — All data stays on your machine as JSON files
• Bilingual — English and German language support
• Recency Boost — Recent knowledge ranks higher
• Usefulness Scoring — Rate entries to improve future recall

Quick Start

Claude Code

claude mcp add memory -- npx mcp-smart-memory

Cursor / Windsurf / Other MCP Clients

Add to your MCP configuration:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["mcp-smart-memory"]
    }
  }
}

That's it. Your agent now has persistent memory.

Tools

| Tool | Description | |------|-------------| | memory_learn | Store knowledge with categories and tags | | memory_recall | Search using TF-IDF semantic similarity | | memory_stats | Knowledge base statistics and overview | | memory_patterns | Analyze patterns, tag frequency, co-occurrence | | memory_suggest | Proactive suggestions based on current context | | memory_evaluate | Rate entries as useful/not useful for better recall |

Usage Examples

Store knowledge

memory_learn: "React useEffect cleanup runs before component unmount and before re-running the effect"
  category: "pattern"
  tags: ["react", "hooks", "cleanup"]

Recall knowledge

memory_recall: "how to clean up effects in React"
→ Returns semantically similar entries ranked by relevance

Get suggestions

memory_suggest: "I'm debugging a memory leak in a React component"
→ Returns relevant warnings, solutions, and best practices from your knowledge base

Configuration

| Environment Variable | Default | Description | |---------------------|---------|-------------| | SMART_MEMORY_DIR | ~/.smart-memory | Data storage directory | | SMART_MEMORY_DB | default | Database name (for multiple knowledge bases) |

Multiple Knowledge Bases

{
  "mcpServers": {
    "work-memory": {
      "command": "npx",
      "args": ["mcp-smart-memory"],
      "env": { "SMART_MEMORY_DB": "work" }
    },
    "personal-memory": {
      "command": "npx",
      "args": ["mcp-smart-memory"],
      "env": { "SMART_MEMORY_DB": "personal" }
    }
  }
}

How It Works

Smart Memory uses TF-IDF (Term Frequency-Inverse Document Frequency) with cosine similarity for semantic search — implemented in pure JavaScript with zero ML dependencies.

When you store knowledge:

1. Text is tokenized and stop-words are removed
2. An inverted index is built for fast lookup
3. TF-IDF vectors are computed for similarity matching

When you search:

1. Your query is vectorized using the same TF-IDF pipeline
2. Cosine similarity is computed against all entries
3. Results are boosted by recency and usefulness ratings
4. Top matches are returned, ranked by relevance score

Data Storage

All data is stored locally as JSON files:

~/.smart-memory/
  └── default/
      ├── knowledge.json    # Your knowledge entries
      └── index.json        # TF-IDF search index

Files are human-readable and portable. Back them up, share them, or version control them.

Why Smart Memory?

| | Smart Memory | claude-mem | mem0 | Zep | |---|---|---|---|---| | MCP Native | ✅ | ❌ Plugin | ❌ SDK | ❌ SDK | | Any MCP Client | ✅ | ❌ Claude only | ❌ Python only | ❌ Python only | | No Cloud | ✅ | ✅ | ❌ | ❌ | | No API Key | ✅ | ✅ | ❌ | ❌ | | Semantic Search | ✅ TF-IDF | ❌ Basic | ✅ Embeddings | ✅ Embeddings | | Pattern Detection | ✅ | ❌ | ❌ | ❌ | | Proactive Suggestions | ✅ | ❌ | ❌ | ❌ |

Built With

Built by NexTool — an AI-native development studio powered by Claude Opus 4.6.

License

MIT