Features

Triggers

Triggers

Watch Supabase signals and workspace metrics, set thresholds, and send alerts to people, webhooks, and agents when they fire.


A trigger watches a series of data, judges it against a condition, and sends an alert to the destinations attached to it when the condition holds. The series can be a signal from a connected Supabase project or any metric in your workspace.

How triggers work

Levels

Triggers can act exactly like alerts and can have a warning level and / or a critical level. A level has a when condition and can have a clear condition. When both levels hold, the trigger goes critical, so a condition that keeps getting worse escalates from warning to critical.

Conditions and events

A condition trigger stays open while its level holds and clears when it stops holding, or when its clear condition holds. An event trigger fires once for each new reading that matches, such as a health check that finds a new priority recommendation.

Missing data

A gap in the data never opens or closes an alert. An open alert stays open until there is a fresh reading to judge. To be told when a database stops answering, use the Database unreachable preset.

When triggers run

Triggers will be caught anytime a stream of data is read into Dreambase. This happens on refreshes for analytics anytime datasets are refreshed, and in observability on a 10 minute schedule.

Default triggers

Dreambase has built out some default Triggers for observability based on recommendations directly from the Supabase team. These are off by default.

Trigger

Signal

Kind

Default level

Database unreachable

Minutes without metrics

Condition

Critical at 10 minutes. Clears after about 10 steady minutes of metrics.

CPU high

CPU usage

Condition

Warning at 80% or more for 10 minutes

Memory high

Memory usage

Condition

Warning at 80% or more for 10 minutes

Disk nearly full

Disk used

Condition

Warning at 90%, critical at 95%. Clears below 87%.

Connections near the limit

Connections in use

Condition

Warning at 80% or more for 5 minutes

Transaction IDs near wraparound

Transaction ID age

Condition

Warning at 1,000 million, critical at 1,400 million

WAL filling the disk

WAL share of disk

Condition

Warning at 25%. Clears below 22%.

Data API pool timeouts

Data API pool timeouts

Condition

Warning at 1 or more timeouts in 20 minutes

Database growing fast

Database size

Condition

Warning at 20% growth in 12 hours

Priority recommendation identified

New priority recommendations

Event

Critical when a check finds one

Recommendation identified

New recommendations

Event

Warning when the hourly check finds one

New recurring peak

New recurring peaks

Event

Warning when the daily check finds one

Health grade drops

Health score

Event

Warning at a 10-point drop since the last check

Turn on alerts for a project

  1. Open Observability and select the project.

  2. Open the Settings tab and go to Triggers. If alerts are off for the whole workspace, click Turn alerts on.

  3. Turn on the project’s switch to turn on every trigger, or turn on only the triggers you want.

  4. Edit any default levels in the rows.

Settings turns defaults on and off but doesn’t delete them. Default triggers can be deleted via the Dreambase MCP utilizing delete_trigger.

Destinations

Add destinations under Notification Destinations on the project’s Settings tab, then attach them to each trigger with the picker on its row. A trigger can send warnings to one destination and critical alerts to another.

  • Email to the workspace owners and admins you pick: one email when an alert starts and one when it clears.

  • Slack through an incoming webhook: one message when an alert starts and one when it clears.

  • Webhook: a signed POST to your endpoint. The signing secret is shown once, when you add it. Events are observability.alert.firing and observability.alert.resolved for a project’s alerts, and trigger.firing and trigger.resolved for a metric’s.

  • Claude routine and Grok Bot: an agent that investigates the alert. These hear default triggers only.

  • ChatGPT and other agents subscribe via MCP Events (below).

Destinations do not automatically attach to a Trigger. These can be set either via the Dreambase MCP (recommended), or by manually editing per Trigger in either the Triggers top level menu, or in Projects -> Settings -> Triggers. Use Test on a destination’s row to check delivery.

MCP Events

Agents connected to the Dreambase MCP subscribe to alerts without a destination:

  • trigger: every alert in the workspace, from Supabase projects and metrics. A token without metrics:read doesn’t hear triggers that read a metric.

  • observability.alert: alerts on Supabase projects only.

Create a trigger through MCP

Custom triggers are created by an agent connected to the Official Dreambase MCP. If the latest Dreambase MCP is installed an agent should be able to pick up the intention and begin creating triggers through normal prompting. Otherwise the following steps can be utilized:

  1. In Triggers, click Add next to Custom. Dreambase copies a prompt for the project.

  2. Paste the prompt into Claude, ChatGPT, Cursor, Codex, or another MCP client.

  3. A natural language description of what to watch is all that is required. An agent can call list_triggers, propose a signal and levels, executes test_trigger, and validates final state before it calls create_trigger.

Custom triggers on a project appear under Custom in Triggers, where they can be turned on or off and destinations can be attached. Change their conditions through the agent.

What a trigger can watch

Each minute, per Supabase project: CPU usage, memory usage, disk used, connections in use, and WAL share of disk (percent); database size (bytes); transaction ID age (millions); Data API pool timeouts (count per minute); minutes without metrics.

At each health check: estimated monthly bill and plan and compute price (USD); monthly active users and total users (count); storage size (bytes); open recommendations and priority recommendations; health score (0 to 100); new priority recommendations, new recommendations, and new recurring peaks.

Metrics: any workspace metric, one reading per scheduled refresh.

Conditions

A condition compares numbers with >, >=, <, and <=, and combines comparisons with &&, ||, and !(…). value is the trigger’s own series, and every level’s when reads it. Windows are written as "10m", "2h", or "1d". A condition is at most 500 characters.

To fire when

Condition

CPU is at or above 80%

value >= 80

It holds for 10 minutes

held(value >= 80, "10m")

CPU and connections are both high, on a CPU trigger

held(value > 85 && connectionsPercent > 80, "10m")

More than 5 pool timeouts happen in 20 minutes

sum(value, "20m") > 5

The database grew 20% in 12 hours

change(value, "12h") >= 20

The score fell 10 points since the last reading

delta(value) <= -10

A signal has had no reading for 30 minutes

silence(value) >= 30

Another project’s CPU is also high

value > 80 && project("<connection-id>").cpuUsage > 80

The series is over 100 times a metric

value / metric("<metric-id>") > 100

Backtest a trigger

test_trigger replays a saved or unsaved trigger over past data and returns how often it would have fired, for how long, and when. Nothing is recorded or sent. A backtest covers the last 7 days by default, and up to 14 days of minute signals, 90 days of health-check signals, and 365 days of metrics. Over more than 3 days, minute signals are judged on hourly averages. A trigger that counts, totals, measures silence, or measures change over less than whole hours is tested over 3 days at most.

Tool reference

list_triggers

MCP and REST · health-reports:read

{ workspaceId, connectionId?, limit?, cursor? }

Lists a workspace’s triggers, oldest first, defaults included. Pass connectionId for one project. Returns at most 100 per page; page until nextCursor is absent.

create_trigger

MCP and REST · health-reports:write

{ workspaceId, name, series, levels, source?, then?, labels?, isEnabled? }

Creates a trigger. series is { kind: "system", signal, connectionId } or { kind: "metric", metricId }. then holds notify steps naming destination ids from get_observability_alerts. A new trigger is on unless isEnabled is false.

update_trigger

MCP and REST · health-reports:write

{ workspaceId, triggerId, name?, levels?, then?, labels?, isEnabled? }

Changes a trigger. levels and then are replaced whole. The series and source can’t change; create a new trigger to watch something else. Turning a trigger off ends its open alert without a notice.

delete_trigger

MCP and REST · health-reports:write

{ workspaceId, triggerId }

Deletes a trigger and its firings. An open alert ends without a notice.

test_trigger

MCP and REST · health-reports:read

{ workspaceId, triggerId? | series + levels, source?, days? }

Backtests a saved trigger, or an unsaved one from its series and levels.

list_trigger_firings

MCP and REST · health-reports:read

{ workspaceId, from?, to?, rule?, connectionId?, limit? }

Lists alerts that started in a window, newest first (the last 7 days by default), plus every alert still open.

Create a trigger through the API

curl -X POST \
  https://api.dreambase.com/v1/workspaces/<workspace-id>/triggers \
  -H "Authorization: Bearer <access-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "CPU sustained high",
    "series": { "kind": "system", "signal": "cpuUsage", "connectionId": "<connection-id>" },
    "levels": [
      { "severity": "warning", "when": "held(value >= 70, \"10m\")", "clear": "value < 60" },
      { "severity": "critical", "when": "held(value >= 90, \"5m\")", "clear": "value < 80" }
    ],
    "then": [
      { "kind": "notify", "channels": ["<slack-destination-id>"], "severities": ["warning"] },
      { "kind": "notify", "channels": ["<email-destination-id>"], "severities": ["critical"] }
    ]
  }'

REST paths: GET and POST /workspaces/{workspace_id}/triggers, PATCH and DELETE /workspaces/{workspace_id}/triggers/{trigger_id}, POST /workspaces/{workspace_id}/triggers/test, GET /workspaces/{workspace_id}/triggers/firings.

Important boundaries

  • Triggers are for workspace owners and admins with Observability on. Metric triggers also need analytics on, metrics:read, and metrics:write to create, change, or delete them.

  • Each Supabase project holds up to 50 custom triggers besides its defaults; each workspace holds up to 50 triggers on metrics.

  • A trigger’s series and source are fixed after creation.

  • Claude routine and Grok Bot destinations hear default triggers only. Custom triggers reach agents through MCP Events or a webhook.

  • Settings doesn’t show metric triggers and can’t delete defaults; use MCP or REST.

  • A trigger that names a project or metric the workspace no longer has stops being judged until its levels change or it is deleted.

  • The trigger tools are marked [EXPERIMENTAL] in their MCP descriptions, so their arguments can change without notice.


A trigger watches a series of data, judges it against a condition, and sends an alert to the destinations attached to it when the condition holds. The series can be a signal from a connected Supabase project or any metric in your workspace.

How triggers work

Levels

Triggers can act exactly like alerts and can have a warning level and / or a critical level. A level has a when condition and can have a clear condition. When both levels hold, the trigger goes critical, so a condition that keeps getting worse escalates from warning to critical.

Conditions and events

A condition trigger stays open while its level holds and clears when it stops holding, or when its clear condition holds. An event trigger fires once for each new reading that matches, such as a health check that finds a new priority recommendation.

Missing data

A gap in the data never opens or closes an alert. An open alert stays open until there is a fresh reading to judge. To be told when a database stops answering, use the Database unreachable preset.

When triggers run

Triggers will be caught anytime a stream of data is read into Dreambase. This happens on refreshes for analytics anytime datasets are refreshed, and in observability on a 10 minute schedule.

Default triggers

Dreambase has built out some default Triggers for observability based on recommendations directly from the Supabase team. These are off by default.

Trigger

Signal

Kind

Default level

Database unreachable

Minutes without metrics

Condition

Critical at 10 minutes. Clears after about 10 steady minutes of metrics.

CPU high

CPU usage

Condition

Warning at 80% or more for 10 minutes

Memory high

Memory usage

Condition

Warning at 80% or more for 10 minutes

Disk nearly full

Disk used

Condition

Warning at 90%, critical at 95%. Clears below 87%.

Connections near the limit

Connections in use

Condition

Warning at 80% or more for 5 minutes

Transaction IDs near wraparound

Transaction ID age

Condition

Warning at 1,000 million, critical at 1,400 million

WAL filling the disk

WAL share of disk

Condition

Warning at 25%. Clears below 22%.

Data API pool timeouts

Data API pool timeouts

Condition

Warning at 1 or more timeouts in 20 minutes

Database growing fast

Database size

Condition

Warning at 20% growth in 12 hours

Priority recommendation identified

New priority recommendations

Event

Critical when a check finds one

Recommendation identified

New recommendations

Event

Warning when the hourly check finds one

New recurring peak

New recurring peaks

Event

Warning when the daily check finds one

Health grade drops

Health score

Event

Warning at a 10-point drop since the last check

Turn on alerts for a project

  1. Open Observability and select the project.

  2. Open the Settings tab and go to Triggers. If alerts are off for the whole workspace, click Turn alerts on.

  3. Turn on the project’s switch to turn on every trigger, or turn on only the triggers you want.

  4. Edit any default levels in the rows.

Settings turns defaults on and off but doesn’t delete them. Default triggers can be deleted via the Dreambase MCP utilizing delete_trigger.

Destinations

Add destinations under Notification Destinations on the project’s Settings tab, then attach them to each trigger with the picker on its row. A trigger can send warnings to one destination and critical alerts to another.

  • Email to the workspace owners and admins you pick: one email when an alert starts and one when it clears.

  • Slack through an incoming webhook: one message when an alert starts and one when it clears.

  • Webhook: a signed POST to your endpoint. The signing secret is shown once, when you add it. Events are observability.alert.firing and observability.alert.resolved for a project’s alerts, and trigger.firing and trigger.resolved for a metric’s.

  • Claude routine and Grok Bot: an agent that investigates the alert. These hear default triggers only.

  • ChatGPT and other agents subscribe via MCP Events (below).

Destinations do not automatically attach to a Trigger. These can be set either via the Dreambase MCP (recommended), or by manually editing per Trigger in either the Triggers top level menu, or in Projects -> Settings -> Triggers. Use Test on a destination’s row to check delivery.

MCP Events

Agents connected to the Dreambase MCP subscribe to alerts without a destination:

  • trigger: every alert in the workspace, from Supabase projects and metrics. A token without metrics:read doesn’t hear triggers that read a metric.

  • observability.alert: alerts on Supabase projects only.

Create a trigger through MCP

Custom triggers are created by an agent connected to the Official Dreambase MCP. If the latest Dreambase MCP is installed an agent should be able to pick up the intention and begin creating triggers through normal prompting. Otherwise the following steps can be utilized:

  1. In Triggers, click Add next to Custom. Dreambase copies a prompt for the project.

  2. Paste the prompt into Claude, ChatGPT, Cursor, Codex, or another MCP client.

  3. A natural language description of what to watch is all that is required. An agent can call list_triggers, propose a signal and levels, executes test_trigger, and validates final state before it calls create_trigger.

Custom triggers on a project appear under Custom in Triggers, where they can be turned on or off and destinations can be attached. Change their conditions through the agent.

What a trigger can watch

Each minute, per Supabase project: CPU usage, memory usage, disk used, connections in use, and WAL share of disk (percent); database size (bytes); transaction ID age (millions); Data API pool timeouts (count per minute); minutes without metrics.

At each health check: estimated monthly bill and plan and compute price (USD); monthly active users and total users (count); storage size (bytes); open recommendations and priority recommendations; health score (0 to 100); new priority recommendations, new recommendations, and new recurring peaks.

Metrics: any workspace metric, one reading per scheduled refresh.

Conditions

A condition compares numbers with >, >=, <, and <=, and combines comparisons with &&, ||, and !(…). value is the trigger’s own series, and every level’s when reads it. Windows are written as "10m", "2h", or "1d". A condition is at most 500 characters.

To fire when

Condition

CPU is at or above 80%

value >= 80

It holds for 10 minutes

held(value >= 80, "10m")

CPU and connections are both high, on a CPU trigger

held(value > 85 && connectionsPercent > 80, "10m")

More than 5 pool timeouts happen in 20 minutes

sum(value, "20m") > 5

The database grew 20% in 12 hours

change(value, "12h") >= 20

The score fell 10 points since the last reading

delta(value) <= -10

A signal has had no reading for 30 minutes

silence(value) >= 30

Another project’s CPU is also high

value > 80 && project("<connection-id>").cpuUsage > 80

The series is over 100 times a metric

value / metric("<metric-id>") > 100

Backtest a trigger

test_trigger replays a saved or unsaved trigger over past data and returns how often it would have fired, for how long, and when. Nothing is recorded or sent. A backtest covers the last 7 days by default, and up to 14 days of minute signals, 90 days of health-check signals, and 365 days of metrics. Over more than 3 days, minute signals are judged on hourly averages. A trigger that counts, totals, measures silence, or measures change over less than whole hours is tested over 3 days at most.

Tool reference

list_triggers

MCP and REST · health-reports:read

{ workspaceId, connectionId?, limit?, cursor? }

Lists a workspace’s triggers, oldest first, defaults included. Pass connectionId for one project. Returns at most 100 per page; page until nextCursor is absent.

create_trigger

MCP and REST · health-reports:write

{ workspaceId, name, series, levels, source?, then?, labels?, isEnabled? }

Creates a trigger. series is { kind: "system", signal, connectionId } or { kind: "metric", metricId }. then holds notify steps naming destination ids from get_observability_alerts. A new trigger is on unless isEnabled is false.

update_trigger

MCP and REST · health-reports:write

{ workspaceId, triggerId, name?, levels?, then?, labels?, isEnabled? }

Changes a trigger. levels and then are replaced whole. The series and source can’t change; create a new trigger to watch something else. Turning a trigger off ends its open alert without a notice.

delete_trigger

MCP and REST · health-reports:write

{ workspaceId, triggerId }

Deletes a trigger and its firings. An open alert ends without a notice.

test_trigger

MCP and REST · health-reports:read

{ workspaceId, triggerId? | series + levels, source?, days? }

Backtests a saved trigger, or an unsaved one from its series and levels.

list_trigger_firings

MCP and REST · health-reports:read

{ workspaceId, from?, to?, rule?, connectionId?, limit? }

Lists alerts that started in a window, newest first (the last 7 days by default), plus every alert still open.

Create a trigger through the API

curl -X POST \
  https://api.dreambase.com/v1/workspaces/<workspace-id>/triggers \
  -H "Authorization: Bearer <access-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "CPU sustained high",
    "series": { "kind": "system", "signal": "cpuUsage", "connectionId": "<connection-id>" },
    "levels": [
      { "severity": "warning", "when": "held(value >= 70, \"10m\")", "clear": "value < 60" },
      { "severity": "critical", "when": "held(value >= 90, \"5m\")", "clear": "value < 80" }
    ],
    "then": [
      { "kind": "notify", "channels": ["<slack-destination-id>"], "severities": ["warning"] },
      { "kind": "notify", "channels": ["<email-destination-id>"], "severities": ["critical"] }
    ]
  }'

REST paths: GET and POST /workspaces/{workspace_id}/triggers, PATCH and DELETE /workspaces/{workspace_id}/triggers/{trigger_id}, POST /workspaces/{workspace_id}/triggers/test, GET /workspaces/{workspace_id}/triggers/firings.

Important boundaries

  • Triggers are for workspace owners and admins with Observability on. Metric triggers also need analytics on, metrics:read, and metrics:write to create, change, or delete them.

  • Each Supabase project holds up to 50 custom triggers besides its defaults; each workspace holds up to 50 triggers on metrics.

  • A trigger’s series and source are fixed after creation.

  • Claude routine and Grok Bot destinations hear default triggers only. Custom triggers reach agents through MCP Events or a webhook.

  • Settings doesn’t show metric triggers and can’t delete defaults; use MCP or REST.

  • A trigger that names a project or metric the workspace no longer has stops being judged until its levels change or it is deleted.

  • The trigger tools are marked [EXPERIMENTAL] in their MCP descriptions, so their arguments can change without notice.


A trigger watches a series of data, judges it against a condition, and sends an alert to the destinations attached to it when the condition holds. The series can be a signal from a connected Supabase project or any metric in your workspace.

How triggers work

Levels

Triggers can act exactly like alerts and can have a warning level and / or a critical level. A level has a when condition and can have a clear condition. When both levels hold, the trigger goes critical, so a condition that keeps getting worse escalates from warning to critical.

Conditions and events

A condition trigger stays open while its level holds and clears when it stops holding, or when its clear condition holds. An event trigger fires once for each new reading that matches, such as a health check that finds a new priority recommendation.

Missing data

A gap in the data never opens or closes an alert. An open alert stays open until there is a fresh reading to judge. To be told when a database stops answering, use the Database unreachable preset.

When triggers run

Triggers will be caught anytime a stream of data is read into Dreambase. This happens on refreshes for analytics anytime datasets are refreshed, and in observability on a 10 minute schedule.

Default triggers

Dreambase has built out some default Triggers for observability based on recommendations directly from the Supabase team. These are off by default.

Trigger

Signal

Kind

Default level

Database unreachable

Minutes without metrics

Condition

Critical at 10 minutes. Clears after about 10 steady minutes of metrics.

CPU high

CPU usage

Condition

Warning at 80% or more for 10 minutes

Memory high

Memory usage

Condition

Warning at 80% or more for 10 minutes

Disk nearly full

Disk used

Condition

Warning at 90%, critical at 95%. Clears below 87%.

Connections near the limit

Connections in use

Condition

Warning at 80% or more for 5 minutes

Transaction IDs near wraparound

Transaction ID age

Condition

Warning at 1,000 million, critical at 1,400 million

WAL filling the disk

WAL share of disk

Condition

Warning at 25%. Clears below 22%.

Data API pool timeouts

Data API pool timeouts

Condition

Warning at 1 or more timeouts in 20 minutes

Database growing fast

Database size

Condition

Warning at 20% growth in 12 hours

Priority recommendation identified

New priority recommendations

Event

Critical when a check finds one

Recommendation identified

New recommendations

Event

Warning when the hourly check finds one

New recurring peak

New recurring peaks

Event

Warning when the daily check finds one

Health grade drops

Health score

Event

Warning at a 10-point drop since the last check

Turn on alerts for a project

  1. Open Observability and select the project.

  2. Open the Settings tab and go to Triggers. If alerts are off for the whole workspace, click Turn alerts on.

  3. Turn on the project’s switch to turn on every trigger, or turn on only the triggers you want.

  4. Edit any default levels in the rows.

Settings turns defaults on and off but doesn’t delete them. Default triggers can be deleted via the Dreambase MCP utilizing delete_trigger.

Destinations

Add destinations under Notification Destinations on the project’s Settings tab, then attach them to each trigger with the picker on its row. A trigger can send warnings to one destination and critical alerts to another.

  • Email to the workspace owners and admins you pick: one email when an alert starts and one when it clears.

  • Slack through an incoming webhook: one message when an alert starts and one when it clears.

  • Webhook: a signed POST to your endpoint. The signing secret is shown once, when you add it. Events are observability.alert.firing and observability.alert.resolved for a project’s alerts, and trigger.firing and trigger.resolved for a metric’s.

  • Claude routine and Grok Bot: an agent that investigates the alert. These hear default triggers only.

  • ChatGPT and other agents subscribe via MCP Events (below).

Destinations do not automatically attach to a Trigger. These can be set either via the Dreambase MCP (recommended), or by manually editing per Trigger in either the Triggers top level menu, or in Projects -> Settings -> Triggers. Use Test on a destination’s row to check delivery.

MCP Events

Agents connected to the Dreambase MCP subscribe to alerts without a destination:

  • trigger: every alert in the workspace, from Supabase projects and metrics. A token without metrics:read doesn’t hear triggers that read a metric.

  • observability.alert: alerts on Supabase projects only.

Create a trigger through MCP

Custom triggers are created by an agent connected to the Official Dreambase MCP. If the latest Dreambase MCP is installed an agent should be able to pick up the intention and begin creating triggers through normal prompting. Otherwise the following steps can be utilized:

  1. In Triggers, click Add next to Custom. Dreambase copies a prompt for the project.

  2. Paste the prompt into Claude, ChatGPT, Cursor, Codex, or another MCP client.

  3. A natural language description of what to watch is all that is required. An agent can call list_triggers, propose a signal and levels, executes test_trigger, and validates final state before it calls create_trigger.

Custom triggers on a project appear under Custom in Triggers, where they can be turned on or off and destinations can be attached. Change their conditions through the agent.

What a trigger can watch

Each minute, per Supabase project: CPU usage, memory usage, disk used, connections in use, and WAL share of disk (percent); database size (bytes); transaction ID age (millions); Data API pool timeouts (count per minute); minutes without metrics.

At each health check: estimated monthly bill and plan and compute price (USD); monthly active users and total users (count); storage size (bytes); open recommendations and priority recommendations; health score (0 to 100); new priority recommendations, new recommendations, and new recurring peaks.

Metrics: any workspace metric, one reading per scheduled refresh.

Conditions

A condition compares numbers with >, >=, <, and <=, and combines comparisons with &&, ||, and !(…). value is the trigger’s own series, and every level’s when reads it. Windows are written as "10m", "2h", or "1d". A condition is at most 500 characters.

To fire when

Condition

CPU is at or above 80%

value >= 80

It holds for 10 minutes

held(value >= 80, "10m")

CPU and connections are both high, on a CPU trigger

held(value > 85 && connectionsPercent > 80, "10m")

More than 5 pool timeouts happen in 20 minutes

sum(value, "20m") > 5

The database grew 20% in 12 hours

change(value, "12h") >= 20

The score fell 10 points since the last reading

delta(value) <= -10

A signal has had no reading for 30 minutes

silence(value) >= 30

Another project’s CPU is also high

value > 80 && project("<connection-id>").cpuUsage > 80

The series is over 100 times a metric

value / metric("<metric-id>") > 100

Backtest a trigger

test_trigger replays a saved or unsaved trigger over past data and returns how often it would have fired, for how long, and when. Nothing is recorded or sent. A backtest covers the last 7 days by default, and up to 14 days of minute signals, 90 days of health-check signals, and 365 days of metrics. Over more than 3 days, minute signals are judged on hourly averages. A trigger that counts, totals, measures silence, or measures change over less than whole hours is tested over 3 days at most.

Tool reference

list_triggers

MCP and REST · health-reports:read

{ workspaceId, connectionId?, limit?, cursor? }

Lists a workspace’s triggers, oldest first, defaults included. Pass connectionId for one project. Returns at most 100 per page; page until nextCursor is absent.

create_trigger

MCP and REST · health-reports:write

{ workspaceId, name, series, levels, source?, then?, labels?, isEnabled? }

Creates a trigger. series is { kind: "system", signal, connectionId } or { kind: "metric", metricId }. then holds notify steps naming destination ids from get_observability_alerts. A new trigger is on unless isEnabled is false.

update_trigger

MCP and REST · health-reports:write

{ workspaceId, triggerId, name?, levels?, then?, labels?, isEnabled? }

Changes a trigger. levels and then are replaced whole. The series and source can’t change; create a new trigger to watch something else. Turning a trigger off ends its open alert without a notice.

delete_trigger

MCP and REST · health-reports:write

{ workspaceId, triggerId }

Deletes a trigger and its firings. An open alert ends without a notice.

test_trigger

MCP and REST · health-reports:read

{ workspaceId, triggerId? | series + levels, source?, days? }

Backtests a saved trigger, or an unsaved one from its series and levels.

list_trigger_firings

MCP and REST · health-reports:read

{ workspaceId, from?, to?, rule?, connectionId?, limit? }

Lists alerts that started in a window, newest first (the last 7 days by default), plus every alert still open.

Create a trigger through the API

curl -X POST \
  https://api.dreambase.com/v1/workspaces/<workspace-id>/triggers \
  -H "Authorization: Bearer <access-token>" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "CPU sustained high",
    "series": { "kind": "system", "signal": "cpuUsage", "connectionId": "<connection-id>" },
    "levels": [
      { "severity": "warning", "when": "held(value >= 70, \"10m\")", "clear": "value < 60" },
      { "severity": "critical", "when": "held(value >= 90, \"5m\")", "clear": "value < 80" }
    ],
    "then": [
      { "kind": "notify", "channels": ["<slack-destination-id>"], "severities": ["warning"] },
      { "kind": "notify", "channels": ["<email-destination-id>"], "severities": ["critical"] }
    ]
  }'

REST paths: GET and POST /workspaces/{workspace_id}/triggers, PATCH and DELETE /workspaces/{workspace_id}/triggers/{trigger_id}, POST /workspaces/{workspace_id}/triggers/test, GET /workspaces/{workspace_id}/triggers/firings.

Important boundaries

  • Triggers are for workspace owners and admins with Observability on. Metric triggers also need analytics on, metrics:read, and metrics:write to create, change, or delete them.

  • Each Supabase project holds up to 50 custom triggers besides its defaults; each workspace holds up to 50 triggers on metrics.

  • A trigger’s series and source are fixed after creation.

  • Claude routine and Grok Bot destinations hear default triggers only. Custom triggers reach agents through MCP Events or a webhook.

  • Settings doesn’t show metric triggers and can’t delete defaults; use MCP or REST.

  • A trigger that names a project or metric the workspace no longer has stops being judged until its levels change or it is deleted.

  • The trigger tools are marked [EXPERIMENTAL] in their MCP descriptions, so their arguments can change without notice.

Team (Members & Roles)

Domains

© 2026 Dream, Inc. All rights reserved.