# Agent-managed settings

Let your agent tune detection and alert thresholds without exposing notification credentials.


## Read before changing

The authenticated project owner can use `get_detection_settings` to inspect the current policy and `update_detection_settings` to change only requested fields. The tools do not return notification destinations or handoff secrets.

```json title="Read the current policy"
{
  "name": "get_detection_settings",
  "arguments": { "projectId": "11111111-1111-4111-8111-111111111111" }
}
```

Replace the example ID with a project returned by `list_projects`. Your MCP client determines how it presents and approves tool calls.

## Detection policy

| Field | Default | Accepted values |
| --- | --- | --- |
| `windowMinutes` | 5 | 1–60 |
| `minOccurrences` | 5 | 2–1,000 |
| `minContexts` | 2 | 1–100 |
| `serverOccurrences` | 10 | 2–1,000 |
| `minRequests` | 0 | 0–100,000 |
| `errorRatePercent` | 0 | 0–100 |

Zero disables the corresponding minimum-request or error-rate gate; it does not turn off all detection. Crash signals follow their immediate handling path. Repeated errors are grouped rather than treated as a new problem per request.

```json title="Change one threshold"
{
  "name": "update_detection_settings",
  "arguments": {
    "projectId": "11111111-1111-4111-8111-111111111111",
    "changes": { "minOccurrences": 8 }
  }
}
```

`reset_detection_settings` restores the recommended defaults. It leaves monitoring schedules, agent handoffs, and notification configuration unchanged.

## Monitor alert policy

First call `list_monitor_alert_settings`. Automatically discovered services are passive and are not editable through `update_monitor_alert_settings`. For an existing manual service or endpoint monitor, the agent can change:

| Field | Meaning |
| --- | --- |
| `enabled` | Whether the monitor reports problems to Inbox |
| `failureThreshold` | Consecutive failed checks; default 3, range 1–20 |
| `recoveryThreshold` | Consecutive successful checks; default 2, range 1–20 |
| `latencyWarningMs` | Slow-response warning, 50–60,000 ms; null disables it |

Changing these rules resets the monitor's check streaks. The tool preserves the URL and schedule. Setting alert `enabled` to false **does not stop HTTP checks**; disable the check itself in its monitor configuration to stop requests.

## What these tools do not change

These tools do not create a monitor, change its target or interval, install an SDK, edit billing, or configure Telegram/SMS/email recipients. Keep those distinctions explicit when asking an agent to “disable monitoring.”

See [endpoint checks](/docs/features/requests), [notifications](/docs/platform/notifications), and [SDK privacy](/docs/features/privacy).