FalkorDB MCP Server

What is FalkorDB MCP Server?

A Model Context Protocol (MCP) server for FalkorDB that enables AI assistants like Claude to query and interact with graph databases using natural language. It allows querying graph data with OpenCypher, creating and managing nodes and relationships, listing multiple graphs, and deleting graphs — all through conversational AI.

How to use FalkorDB MCP Server?

Install via npm (@falkordb/mcpserver) and configure environment variables for your FalkorDB instance (FALKORDB_HOST, FALKORDB_PORT, etc.) in a .env file or inline. Add the server to your Claude Desktop config under mcpServers, or run directly with npx using either stdio (default) or HTTP transport. For HTTP transport, set MCP_TRANSPORT=http and optionally MCP_PORT and MCP_API_KEY. A Docker Compose setup is also provided.

Key features of FalkorDB MCP Server

• Query graph databases using OpenCypher (with read-only mode)
• Parameterized queries to prevent injection risks
• Dedicated read-only query tool (query_graph_readonly)
• Create, manage, and delete nodes and relationships
• List available graphs and explore schema (node labels, relationship types)
• Schema discovery tools with sampling and connection topology
• Works with stdio (default) or streamable HTTP transport
• API key authentication for HTTP mode (optional)
• Read-only mode for replica instances (set via FALKORDB_DEFAULT_READONLY)

Use cases of FalkorDB MCP Server

• Ask an AI assistant to retrieve graph data using natural language queries
• Let agents create and update nodes and relationships in a knowledge graph
• Explore graph schemas and property distributions before writing queries
• Run safe, read-only queries on replica or production databases
• Integrate FalkorDB with Claude Desktop for conversational database management

FAQ from FalkorDB MCP Server

What are the prerequisites for using the server?

You need Node.js 18+, a running FalkorDB instance (local or remote), and optionally the Claude Desktop app for AI integration.

Does the server support read-only queries?

Yes. You can set FALKORDB_DEFAULT_READONLY=true for the whole server, or use the readOnly parameter per query. There is also a dedicated query_graph_readonly tool that always runs GRAPH.RO_QUERY.

How do I connect via HTTP transport?

Set MCP_TRANSPORT=http and optionally MCP_PORT (default 3000). Clients send a POST request with an initialize message; the server returns an Mcp-Session-Id header to include in subsequent requests. If MCP_API_KEY is set, clients must provide Authorization: Bearer <key>.

Can I use parameterized queries?

Yes. Both query_graph and query_graph_readonly accept an optional params object. Parameters are referenced as $name in the query text, avoiding injection risks. Note that LIMIT and SKIP clauses do not support parameters.

What schema discovery tools are available?

Three tools help agents orient themselves: get_graph_schema (returns node labels, relationship types, and connection topology), get_node_schema, and get_relationship_schema (both sample properties by frequency). All are read-only (GRAPH.RO_QUERY).