GraphRAG Configuration Guide - Graphwise GraphDB Integration (MCP)

06/07/2026

GraphRAG version 1.2.0

GraphDB provides multi-hop knowledge graph querying tools to the orchestrator layer through the Model Context Protocol (MCP) framework. It offers specialized tools, such as SPARQL creation, ontological mapping, and semantic vector indexing, directly to the AI agent via Server-Sent Events (SSE).

Assign the target integration variables inside your centralized Configuration workflow runtime code block:

graphDBMcpUrl
The target SSE endpoint of your GraphDB Model Context Protocol gateway (e.g., http://graphdb:7200/mcp/sse).
graphDBMcpRepository
The main repository namespace identifier containing your enterprise semantic triples and graph database.
graphDBMcpOntologyGraph
The named graph URI holding your domain model, class structures, properties, and ontological constraints.
graphDBMcpFtsSearchTriplesLimit
Caps the maximum number of matched literal database triple structures returned during full-text search matching (Default: 100).
mcpLLMModel
The specific language model dedicated strictly to building syntactically valid SPARQL statements, validating graph pathways, and translating triple outputs (e.g., anthropic.claude-3-5-sonnet-20240620-v1:0).
// GraphDB MCP endpoint (SSE)
graphDBMcpUrl: "http://graphdb:7200/mcp/sse",
graphDBMcpRepository: "enterprise-knowledge-base",
graphDBMcpOntologyGraph: "[http://example.com/ontology/graphwise](http://example.com/ontology/graphwise)",
graphDBMcpFtsSearchTriplesLimit: 100,
// Dedicated SPARQL generation engine
mcpLLMModel: "anthropic.claude-3-5-sonnet-20240620-v1:0",

The underlying MCP bridge registers a collection of tools that the orchestrator's agent uses to navigate the knowledge graph:

Tool Name	Tool Function / Core Capability
`sparql_query`	Translates natural language queries and executes raw structured SPARQL operations directly against the active repository.
`fts_search`	Performs Lucene-based text searches across registered graph literal values and string indexes.
`ontology_schema`	Returns the structural class model and relationships to assist the agent in drawing valid navigation paths.
`autocomplete_iri_discovery_search`	Uses internal autocomplete indices to rapidly resolve simple terms to valid, unique IRIs.
`similarity_search`	Performs semantic similarity searches on embedded literal values registered inside the graph engine.
`retrieval_search`	Carries out direct retrieval-based node lookups to resolve entity namespaces.
`iri_discovery`	Discovers valid entity IRIs corresponding to conceptual queries inside the operational repository.

To configure access policies for the GraphDB server endpoint (Basic Auth, API Key, or None), follow these steps:

Open your Orchestrator workspace dashboard and locate the MCP Client node within your active GraphDB MCP workflow.
Define your target authentication mechanism within the node properties pane.
Ensure corresponding credentials are saved inside the workflows sidebar to keep authentication headers out of unencrypted log traces.

Caution

Configuration Guardrail

Modifying or reducing the tools array parameters requires a corresponding system prompt alignment inside the GraphDB MCP workflow. The agent's underlying system prompts must explicitly understand changed tool execution boundaries to prevent generation or execution failures.

Tip

Consult the GraphDB LLM Tools section for in-depth technical specifications regarding tool execution schemas and performance optimization.

In this section: