Register a device or group (by explicit address list, or by name pattern / service UUIDs discovery filter) into a named fleet. risk_tier is one of read_only, low_risk_actuator, or safety_critical; safety_critical devices are excluded from fleet_write batches unless explicitly confirmed.

Discover devices matching a filter, without connecting to or registering them. Use this to see what's out there before fleet_register.

Read one resource across every device in a fleet, or a subset by address. Returns per-device results (success / unreachable / timeout / still_queued) as an operation snapshot keyed by operation_id.

Write one resource across a fleet. Requires the server to have been started with FLEET_ALLOW_WRITES=1. Every write is verified with a readback; the result reports both acknowledged and converged. Devices tiered safety_critical are skipped (status confirmation_required) unless their address is passed in confirm_addresses.

Subscribe to (action="subscribe"), stop watching (action="unsubscribe"), or drain buffered notifications for (action="poll") a resource across a fleet. Watching a device pins one pool connection slot for as long as it's watched.

Health of every device in a fleet: healthy/unhealthy, consecutive failure count, and whether it's currently connected.

Connection pool telemetry: active/idle/evicting connection counts, queue depth, and per-device wait times. The live debugging surface for the pool.

Poll a fleet_read/fleet_write batch by operation_id for partial or complete results. Never returns a silently-dropped device: every device resolves to success, a specific error, or still_queued.