# Knowledge Graph RDBMS MCP server

> Provides an embedded label property graph using SQLite or Postgres that allows AI agents to read, write, and reason over structured entity relationships.

- Repository: https://github.com/cunicopia-dev/knowledge-graph-rdbms
- Page: https://hostedmcp.org/servers/cunicopia-dev/knowledge-graph-rdbms
- Transport: stdio
- Stars: 1
- License: MIT
- Latest version: 0.1.9

## Install

### Claude Code

```sh
claude mcp add knowledge-graph-rdbms -- uvx knowledge-graph-rdbms
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "knowledge-graph-rdbms": {
      "command": "uvx",
      "args": [
        "knowledge-graph-rdbms"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "knowledge-graph-rdbms": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "knowledge-graph-rdbms"
      ]
    }
  }
}
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "knowledge-graph-rdbms": {
      "command": "uvx",
      "args": [
        "knowledge-graph-rdbms"
      ]
    }
  }
}
```

### Codex (`~/.codex/config.toml`)

```toml
[mcp_servers.knowledge-graph-rdbms]
command = "uvx"
args = ["knowledge-graph-rdbms"]
```

### Gemini CLI (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "knowledge-graph-rdbms": {
      "command": "uvx",
      "args": [
        "knowledge-graph-rdbms"
      ]
    }
  }
}
```

## Environment variables

| Name | Required | Secret | Description |
|---|---|---|---|
| `KGRDBMS_DEFAULT_ONTOLOGY` | no | no | Name of the default ontology used when a tool call omits the `ontology` argument. Defaults to "default". |
| `KGRDBMS_HOME` | no | no | Storage root for ontology databases and the event log. Defaults to ~/.kgrdbms. |
| `KGRDBMS_MCP_TOKEN` | no | no |  |

## FAQ

### What kind of graph model does it use?

It uses a label property graph model consisting of nodes, typed edges, labels, and JSON properties.

### Can I roll back changes made by an AI agent?

Yes, every write is logged in an append-only event log, allowing you to revert specific events or replay the graph to a specific point in time.

### How does it compare to Neo4j?

Neo4j is better for deep traversals and complex pattern matching in clustered environments, while this tool is designed for small, embedded, local-first graph storage and agent memory.