MCP server for Rachio smart irrigation controllers via the public cloud API.
Rachio MCP has clear, action-oriented tool names (list_*, get_*, start_*, stop_*, set_*) and reasonable descriptions (avg ~120 chars). All 10 tools are explicitly registered with @mcp.tool() decorators. However, input schemas are inferred from function signatures rather than explicitly declared in JSON Schema format, the code shows parameter names and types in Python but no formal schema objects. Parameter descriptions exist in docstrings but are minimal (avg ~40 chars). Output schemas are not documented. Error handling is basic (RuntimeError/ValueError with messages). No tool annotations (readOnlyHint/destructiveHint). The server accepts human-friendly identifiers (device name, zone name) alongside IDs, which is good for usability. Missing: pagination for list_devices/list_zones/list_schedules, explicit output schema documentation, and structured error recovery guidance.
Current state of a controller: online, on/paused, rain sensor/delay, and what (if anything) is watering right now. The one call for "what's my sprinkler doing".
List Rachio controllers on the account with status and zone summary.
List the stored schedule rules (watering times) on a controller.
List every zone on a controller (number, name, enabled state).
Start a stored schedule rule by its id or name.
Turn a controller's automatic schedules on (standby off) or off (standby on). Turning it off pauses all scheduled watering until turned back on.
Input schemas are inferred from Python function signatures, not explicitly declared as JSON Schema objects. LLM clients cannot parse formal constraints (enums, min/max, patterns) from the code.
Output schemas are not documented. Callers cannot know what fields to expect (e.g., does list_devices return 'zones' or 'zone_list'?). This forces LLMs to guess field names for downstream tool calls.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 54 | 2026-07-28+ | v2 |
Set a rain delay in days (0 clears it, max 7).
Run a single zone for the given number of minutes (max 180).
Run multiple zones in sequence. Each item: {"zone": <num/name>, "minutes": <n>}. They water one after another in the order given.
Stop all watering on a controller immediately (manual or scheduled).
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Clients cannot distinguish safe read-only tools from destructive ones without parsing descriptions.
List tools (list_devices, list_zones, list_schedules) lack pagination parameters (limit, offset, cursor). If a user has many devices/zones/schedules, results could exceed context window.
Error messages are generic (RuntimeError, ValueError). No recovery guidance. E.g., 'Device not found' should suggest 'Call list_devices() to see available devices.'