Knowledge Graph MCP Server

How to use

You will run the MCP server as a local process and connect your MCP client (Cursor or Claude Desktop) to it. The server exposes a set of graph, node, edge, and resource management commands that you can invoke from your client to create graphs, add nodes and edges, attach resources, and publish or archive graphs. Use the client’s MCP integration to trigger prompts that query and summarize knowledge graph data, then incorporate those results into your conversations.

How to install

Prerequisites you need on your system before installation are Node.js and pnpm.

# Ensure Node.js is installed (version 16.0.0 or newer)
node -v

# Ensure pnpm is installed (version 7.0.0 or newer)
pnpm -v

Additional setup and usage notes

Create a directory to store knowledge graph data. This directory is referenced by the server as its storage location.

mkdir ~/knowledge_graph

Configuration examples

Configure the MCP server in Cursor to run the MCP client you are using. Save this as part of your Cursor config.

{
  "mcpServers": {
    "knowledge-graph": {
      "command": "npx",
      "args": [
        "-y",
        "@aiuluna/knowledge-graph-mcp",
        "--client",
        "claude"
      ],
      "env": {
        "KNOWLEDGE_GRAPH_DIR": "/path/to/your/knowledge_graph/dir"
      }
    }
  }
}

Usage in Claude Desktop

If you use Claude Desktop, add this configuration to your claude_desktop_config.json so Claude can connect to the MCP server.

{
  "mcpServers": {
    "knowledge-graph": {
      "command": "npx",
      "args": [
        "-y",
        "@aiuluna/knowledge-graph-mcp"
      ],
      "env": {
        "KNOWLEDGE_GRAPH_DIR": "/path/to/your/knowledge_graph/dir"
      }
    }
  }
}

Prompt usage and agent mode

Prompts are used to query the knowledge graph and generate summaries that merge with your ongoing conversation. In Agent mode, trigger knowledge graph queries with the designated command (for example, a slash command or editor shortcut) to have the agent analyze the current context and retrieve relevant graph nodes.

Graph, node, edge, and resource management

You can manage graphs, nodes, edges, and resources through the MCP tool list. This includes creating graphs, listing graphs with optional status and type filters, publishing graphs, adding and updating nodes and edges, and managing resources like SVGs and Markdown files associated with nodes.

Project structure and prompts

Prompt files reside under the prompts directory and should be copied to your client’s expected location. Use the graph-query prompt to drive knowledge graph queries from your agent.

Error handling and troubleshooting

The service provides clear error messages and handling guidance for common issues. If an error occurs, check the log files and verify that the storage directory exists and is accessible by the MCP server.

Development and testing notes

Development tasks for the MCP server use standard tooling: install dependencies, run in development mode, build the project, run tests, and perform linting.

create_graph

Create a new knowledge graph with a name, description, and type (topology, timeline, changelog, requirement, kb, ontology).

list_graphs

List all graphs with optional filters for status and type.

publish_graph

Publish a graph by its Graph ID, changing its status from draft to published.

add_node

Add a node to a specified graph with type, name, and optional description, file path, and metadata.

update_node

Update node attributes such as name, description, file path, or metadata for a given node in a graph.

delete_node

Delete a node from a graph with a delete confirmation flag.

get_node_details

Retrieve detailed information about a specific node within a graph.

add_edge

Add an edge between two nodes with a type, label, weight, and optional metadata.

update_edge

Update edge attributes like label, weight, or metadata.

delete_edge

Remove an edge from a graph with a delete confirmation flag.

get_creation_guidelines

Fetch guidelines for creating resources such as SVG and Markdown for graph nodes.

save_resource

Save a resource (svg or markdown) associated with a graph or node.

update_resource

Update an existing resource’s metadata or title.

delete_resource

Remove a resource with a delete confirmation flag.

unlink_resource

Unlink a resource from a specific node in a graph.