Use KMGraph in Cursor, Windsurf, VS Code, and JetBrains

"I use Cursor/VS Code/Windsurf. How do I get KMGraph working there?"

You need Node.js 18+, the KMGraph repo cloned locally, and your target IDE installed with MCP support.

Step 1 — Build the MCP server

cd /path/to/knowledge-graph/mcp-server
npm install
npm run build

The server binary is at mcp-server/dist/index.js.

Step 2 — Register the MCP server in your IDE

Cursor

Open .cursor/mcp.json (create if absent):

{
  "mcpServers": {
    "kmgraph": {
      "command": "node",
      "args": ["/path/to/knowledge-graph/mcp-server/dist/index.js"],
      "env": {
        "KG_CONFIG_PATH": "/path/to/kg-config.json"
      }
    }
  }
}

Restart Cursor. The kg_* tools appear in the MCP tools panel.

Windsurf

Open .windsurf/mcp.json:

{
  "mcpServers": {
    "kmgraph": {
      "command": "node",
      "args": ["/path/to/knowledge-graph/mcp-server/dist/index.js"]
    }
  }
}

VS Code

Run the VS Code installer script:

bash scripts/install-vscode.sh

The script auto-detects the VS Code settings path and registers the MCP server.

Alternatively, add to .vscode/settings.json:

{
  "mcp.servers": {
    "kmgraph": {
      "command": "node",
      "args": ["/path/to/knowledge-graph/mcp-server/dist/index.js"]
    }
  }
}

JetBrains

In JetBrains settings → Tools → AI Assistant → MCP Servers, add a new server with name kmgraph, command node, and args /path/to/knowledge-graph/mcp-server/dist/index.js.

Continue.dev

In .continue/config.json:

{
  "mcpServers": [
    {
      "name": "kmgraph",
      "command": "node",
      "args": ["/path/to/knowledge-graph/mcp-server/dist/index.js"]
    }
  ]
}

Step 3 — Initialize the knowledge graph

In the IDE's AI chat, ask the assistant:

Use kg_config_init to initialize a new knowledge graph for this project.

Or paste the contents of INSTALL.md for automated setup.

Step 4 — Add workflow instructions

Copy core/default-templates/AGENTS-template.md to your project as AGENTS.md. This file gives the AI assistant the KMGraph workflow instructions for non-Claude-Code platforms.

Available MCP tools

All kg_* tools are available: kg_capture, kg_search, kg_resolve, kg_config_init, kg_config_list, kg_fts5_rebuild, kg_scaffold, kg_check_sensitive.

Ask the AI assistant:

Use kg_search to search for "test" in the knowledge graph.

The tool should return results (or an empty result if the graph is new).

Related

• Platform Adaptation reference — full IDE compatibility matrix
• Migrate between Claude and Gemini — switching platforms with KG continuity