NornicDB

What NornicDB MCP server does

The NornicDB MCP server connects an MCP-capable agent to NornicDB, a database designed to keep graph, vector, and historical data in one system. The database supports graph traversal, vector retrieval, hybrid retrieval, and historical reads backed by multi-version concurrency control (MVCC).

NornicDB is compatible with Neo4j’s Bolt protocol and Cypher query language, so applications that already use those interfaces may be able to reuse existing query patterns and drivers. It also provides REST, GraphQL, and gRPC interfaces, including a Qdrant-compatible gRPC path for client workflows that need vector-database compatibility.

This fits agent memory, Graph-RAG, knowledge systems, and canonical truth stores where relationships and semantic similarity need to be queried together. Its temporal features also support as-of reads, retained history, and audit-oriented retrieval.

How it works

The NornicDB MCP server acts as the agent-facing layer for a NornicDB deployment. The database itself stores graph and vector information together and exposes database access through several protocols. The README describes Bolt and Cypher as the primary Neo4j-compatible access path, while REST, GraphQL, and gRPC provide additional integration options.

Transactions use snapshot isolation. A transaction reads from a specific MVCC version, sees its own buffered writes, and does not automatically see commits made after its snapshot. Conflicting concurrent graph mutations can fail at commit rather than silently replacing newer state. Historical reads are subject to the database’s retained MVCC floor, while current search paths remain focused on current state.

Setup and configuration

The README documents deployment of NornicDB through Homebrew or Docker. Docker examples expose ports 7474 and 7687 and persist data in a mounted /data volume. Images are provided for Apple Silicon with Metal, amd64 CPU hosts, NVIDIA CUDA hosts, and Vulkan hosts. The admin UI is available at http://localhost:7474 after startup.

The provided material does not include MCP-specific installation commands, transport settings, authentication options, or environment variables. It also notes that Docker on macOS does not provide Metal acceleration; native installation or a local build is required for GPU acceleration there.

Tools and capabilities

The NornicDB MCP server is relevant to agents that need access to:

• Cypher queries and Neo4j-compatible Bolt connectivity
• Graph traversal and relationship-oriented retrieval
• Vector and hybrid search
• Temporal facts, MVCC-backed historical reads, and replayable history
• Agent memory, Graph-RAG, schemas, and managed embeddings
• GPU execution paths using CUDA, Metal, or Vulkan where supported

The README references agent-ready skill files covering Cypher query shapes, decay and promotion policies, managed embeddings, vector and hybrid search, and RAG procedures. It does not provide a definitive list of MCP tool names.

Limitations and notes

NornicDB’s documented search paths focus on current state rather than historical MVCC snapshots. Historical data is only available within the retained version range; reads below that floor can return ErrNotFound. The benchmark and retrieval figures in the README are configuration-specific and should not be treated as universal performance guarantees.

The supplied material does not identify supported MCP clients, describe MCP authentication, or state whether the MCP interface uses stdio or HTTP. Developers should verify those details in the repository before configuring an agent client.