Runtime Quick Deployment

1. Start the Launcher locally

The Launcher is the shortest local validation path and requires Java 17 or newer. The example pins v0.1.17; check the project release page before using it in a new environment.

mkdir foggy-runtime && cd foggy-runtime

curl -fLO https://github.com/foggy-projects/foggy-data-mcp-bridge/releases/download/foggy-runtime-launcher-v0.1.17/foggy-runtime-launcher-0.1.17.jar
curl -fLO https://github.com/foggy-projects/foggy-data-mcp-bridge/releases/download/foggy-runtime-launcher-v0.1.17/start-foggy-runtime.sh
curl -fLO https://github.com/foggy-projects/foggy-data-mcp-bridge/releases/download/foggy-runtime-launcher-v0.1.17/SHA256SUMS

grep -E 'foggy-runtime-launcher-0.1.17.jar|start-foggy-runtime.sh' SHA256SUMS | sha256sum -c -
chmod +x start-foggy-runtime.sh
./start-foggy-runtime.sh

Windows users can download start-foggy-runtime.ps1 from the same release. The default URL is http://127.0.0.1:18066, with SQLite as the default local database.

The default Launcher security mode is for development/test use. Do not expose it to the public internet or a production network without management authentication, business identity propagation, network controls, datasource secret protection, and auditing.

2. Probe the service

curl http://127.0.0.1:18066/readyz
curl http://127.0.0.1:18066/api/v1/capabilities

Use the capability response as the automation baseline. Record the Runtime API version, enabled features, and securityMode. If the deployment reports auth-code, Runtime API management calls must include X-Foggy-Runtime-Code.

For a new custom namespace, continue with Runtime API and MCP Examples to test the datasource, bind the namespace, validate the model, and refresh it; readiness alone does not make a model queryable.

3. Connect an MCP client

{
  "mcpServers": {
    "foggy-ai-analysis": {
      "url": "http://127.0.0.1:18066/mcp/analyst/rpc",
      "headers": {
        "X-NS": "salesdrop"
      }
    }
  }
}

You can discover tools directly first:

curl -X POST http://127.0.0.1:18066/mcp/analyst/rpc \
  -H 'Content-Type: application/json' \
  -H 'X-NS: salesdrop' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

X-NS is the isolation axis. The client, Runtime API calls, and model resources must use the same namespace. Keep datasource passwords and management codes out of MCP client configuration.

4. Source integration boundary

The current source uses Java 17, Spring Boot 3.4.x, and com.foggysource coordinates; this manual uses the published 9.2.0 version. For customization, use the bridge repository's pom/BOM as the source-build contract.

For customization, build and run focused checks in the bridge repository before consuming the output from another application. Do not copy the historical 8.1.10.beta snippets into a new integration.

5. AI and chart dependencies

• Structured MCP tools can run without a ChatModel/AI Provider.
• dataset_nl.query requires a compatible AI Provider; without one, use dataset.query_model or dataset.compose_script.
• dataset.export_with_xchart is JVM-native and does not require an external render service.
• dataset.export_with_echarts requires an available ECharts render service; do not use the retired dataset.export_with_chart name.

6. First delivery gate

After startup, capability discovery, namespace selection, datasource binding, model validate/refresh, and one tools/list response, move to query troubleshooting. Preserve request/trace IDs, namespace, generation, and error codes at every stage so startup, model, and query failures remain distinguishable.