ainative-memory-mcp

Enhanced MCP knowledge graph memory server with cloud persistence and semantic search. Drop-in replacement for @modelcontextprotocol/server-memory.

Why this instead of server-memory?

| Feature | server-memory | ainative-memory-mcp | |---------|--------------|-------------------| | Storage | Local JSONL file | ZeroDB cloud | | Persistence | Lost on machine wipe | Survives forever | | Cross-device | No | Yes (same API key = same graph) | | Semantic search | No (text match only) | Yes (vector similarity) | | Setup | Manual file path | Auto-provisions on first run | | Sharing | Copy files around | Share API key |

Quick Start

Claude Desktop / Claude Code

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "ainative-memory-mcp"],
      "env": {
        "ZERODB_API_KEY": "ak_your_key"
      }
    }
  }
}

No API key? Just omit it — a free ZeroDB instance is auto-provisioned on first run.

Cursor / Windsurf / VS Code

Same config in your .mcp.json:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "ainative-memory-mcp"],
      "env": {
        "ZERODB_API_KEY": "ak_your_key"
      }
    }
  }
}

Get an API Key

npx zerodb-cli init

Or sign up at ainative.studio.

Tools (10)

All 9 original server-memory tools plus semantic search:

Original Tools

| Tool | Description | |------|-------------| | create_entities | Create new entities with name, type, and observations | | create_relations | Create directed relations between entities | | add_observations | Add observations to existing entities | | delete_entities | Delete entities and their cascading relations | | delete_observations | Remove specific observations from entities | | delete_relations | Delete specific relations | | read_graph | Read the entire knowledge graph | | search_nodes | Search by text match (name, type, observations) | | open_nodes | Fetch specific entities by name with their relations |

New Tool

| Tool | Description | |------|-------------| | search_nodes_semantic | Vector similarity search — find entities by meaning, not just exact text |

Semantic Search Example

The original search_nodes only finds exact text matches. search_nodes_semantic understands meaning:

search_nodes("ML frameworks")        -> might find nothing
search_nodes_semantic("ML frameworks") -> finds "PyTorch", "TensorFlow", "JAX"

Environment Variables

| Variable | Required | Description | |----------|----------|-------------| | ZERODB_API_KEY | No* | ZeroDB API key (auto-provisions if missing) | | ZERODB_PROJECT_ID | No | ZeroDB project ID | | ZERODB_API_URL | No | API URL (default: https://api.ainative.studio) |

*If no credentials are provided, a free instance is automatically provisioned on first run.

Auto-Provisioning

On first run without credentials:

1. A free ZeroDB instance is provisioned automatically
2. Credentials are saved to .mcp.json and .env in your project
3. A claim URL is printed so you can take ownership of the instance
4. Everything works immediately — no signup required

How It Works

Instead of writing to a local memory.jsonl file, this server:

1. Entities are stored as rows in a ZeroDB NoSQL table (kg_entities)
2. Relations are stored in a separate table (kg_relations)
3. Vector embeddings are automatically generated for each entity via ZeroMemory, enabling semantic search
4. All data persists in the cloud and is accessible from any device with the same API key

Migration from server-memory

1. Replace @modelcontextprotocol/server-memory with ainative-memory-mcp in your MCP config
2. Add ZERODB_API_KEY to your env (or let it auto-provision)
3. Re-create your entities (one-time migration — your old JSONL data stays on disk)

License

MIT