Knowledge Graph Brain exposes its capabilities as MCP tools and REST/OpenAPI endpoints to build, query, and manage a hybrid graph-and-embedding knowledge store. The MCP surface includes tools for knowledge access (ask_knowledge_graph, search_semantic, explore_relationships), lifecycle management (list_knowledge_bases, add_data_source, start_ingestion), and schema exploration (explore_schema, find_patterns, get_overview). You can interact with these tools via MCP clients or expose them through REST/OpenAPI for integration with apps like Open WebUI, Claude Desktop, or VS Code MCP extensions. To get started, run the MCP server, register your data sources and schemas, ingest data, and then query the graph with natural-language or structured prompts. The included web UI and OpenAPI docs help you explore capabilities and test tool calls during development. For local experimentation, you can build the MCP server and run it with mcpo, then access the generated docs at the provided endpoint.

• Prerequisites:

  • Node.js and npm installed on your machine
  • Neo4j database running (default bolt://localhost:7687 with user neo4j/password or your configured credentials)
  • Git available to clone the repository

• Installation steps:

  1. Clone the repository git clone https://github.com/ryandmonk/knowledge_graph_brain.git cd knowledge_graph_brain

  2. Install dependencies npm install

  3. Build/prepare the MCP server package

     From the repository root, build the MCP server (examples show in the docs)

     cd mcp-server && npm run build

  4. Run the MCP server via mcpo (example)

     From the repo root after building, or adjust paths as needed

     ../.venv/bin/mcpo --port 8080 -- node ./dist/index.js

  5. Open the API/docs endpoint to verify REST/OpenAPI surface open http://localhost:8080/docs

• Notes:

  • If you use a custom Neo4j setup, set NEO4J_URI, NEO4J_USER, and NEO4J_PASSWORD in the environment or in the config file.
  • The project includes a setup wizard and UI for local exploration; ensure you have the orchestrator UI scripts running if you want the local UI features.

Tips and gotchas:

• The MCP surface exposes both tool calls and REST/OpenAPI endpoints; you can mix MCP tool invocations with standard HTTP requests for testing and integration.
• Ensure Neo4j is up before starting ingestion or queries that rely on the graph data.
• When debugging, use the provided UI features to inspect ingestion status, schema configurations, and service visibility (Neo4j, connectors).
• The Quick Start workflow demonstrates cloning, installing, starting services, and ingesting data; adapt the commands to your environment if you host services differently.
• Environment variables can be used to customize endpoints and credentials; document them in your deployment config for reproducibility.