CentralMind Gateway

Gateway uses an LLM only during this one-time step to analyze your schema and produce a gateway.yaml configuration file. The example uses Gemini (free tier). Set GEMINI_API_KEY first, then point the tool at your database.

export GEMINI_API_KEY='your-gemini-api-key'

docker run --platform linux/amd64 -it \
  -e GEMINI_API_KEY=$GEMINI_API_KEY \
  -v $(pwd):/workspace \
  ghcr.io/centralmind/gateway:v0.2.18 discover \
  --ai-provider gemini \
  --connection-string "postgresql://user:password@host:5432/mydb?sslmode=require" \
  --prompt "Generate a readonly API for analytics" \
  --output /workspace/gateway.yaml

The discovery step creates gateway.yaml listing all discovered tables, columns, and generated API endpoints. Review it to confirm no sensitive columns are exposed and that the generated endpoint paths match your expectations before starting the server.

Run the gateway with your generated config. It exposes an MCP SSE endpoint at /sse and a Swagger UI at the root. The LLM API key is NOT needed at runtime — inference only happened during discovery.

docker run --platform linux/amd64 -p 9090:9090 \
  -v $(pwd)/gateway.yaml:/gateway.yaml \
  ghcr.io/centralmind/gateway:v0.2.18 start \
  --config /gateway.yaml stdio

Point Claude Desktop or Cursor at the Gateway SSE endpoint. The MCP server is now live and all discovered database endpoints appear as tools.

{
  "mcpServers": {
    "centralmind-gateway": {
      "command": "docker",
      "args": [
        "run", "-i", "--platform", "linux/amd64",
        "ghcr.io/centralmind/gateway:v0.2.18",
        "--connection-string", "postgresql://user:password@host:5432/mydb",
        "start", "stdio"
      ]
    }
  }
}

Add plugin entries to gateway.yaml to activate PII scrubbing (Microsoft Presidio or regex), API key auth, OAuth, or OpenTelemetry tracing. See the plugin documentation at docs.centralmind.ai/plugins for YAML syntax.