--- name: relay-setup description: Connect Relay for app testing, diagnose an unreachable service or incompatible tools, and get to a first verdict. --- # Connect Relay If the host exposes no Relay tools, follow the plugin README to configure the installed connector executable (the plugin passes `--profile device`; without a profile it uses `qa`, which cannot drive devices or record) with the intended service or workspace. Its `--help` works offline. 1. Call `relay_get_guide` for `start` and `agents`. Guides ship with the connector and stay readable while the service is offline. 2. Call `relay_health`. If the service is unavailable, run `relay-mcp doctor --profile device` in the host's environment and follow its failed check's next action. With the matching `@relay/runtime` installed, an explicit absolute `--workspace` attaches to or launches the local service (`--runtime-port` when the default is taken). An explicit `--server` or `RELAY_URL` keeps that endpoint. Doctor never launches. 3. Call `relay_list_apps`, `relay_list_tests` and `relay_list_runs` (or `relay_panel`) to see Apps, Tests and recent Runs. 4. Call `relay_list_devices`; pick a ready target explicitly when several exist and keep its `targetId`. 5. First verdict: `relay_create_test` with a sentence and the `url` or `app`, then the returned `relay_run_test` call. Running steps written from words needs a model key (`OPENROUTER_API_KEY` or one saved in Settings); without one, record the Test instead (`relay-debug-and-record`). Credentials belong in the host process environment. Use a distinct `agent:` actor. Another actor's control or an active Run means waiting or authorized cancellation. A healthy connection alone is not a tested app.