The Monte Carlo Agent Toolkit now includes instrument-agent, a skill that automates your agent-side setup for Monte Carlo Agent Observability.

From a single prompt in your editor, the skill detects the AI libraries in use, installs the Monte Carlo OpenTelemetry SDK at compatible versions, places decorators that categorize LLM calls for evaluation, and verifies traces are landing in Monte Carlo.

Available in Claude Code, Cursor, VS Code, Copilot CLI, and Codex -- wherever you build. Install the Monte Carlo Agent Toolkit and run /instrument-agent in your agent codebase.

Learn more:

• https://docs.getmontecarlo.com/docs/agent-toolkit
  • https://github.com/monte-carlo-data/mc-agent-toolkit/blob/main/skills/instrument-agent/SKILL.md

The Alerts page now includes a Triage alerts banner. With one click, an AI agent runs through selected alerts (up to 50, respecting your current filters) and prioritizes the alert feed into high, medium, and low confidence buckets.

This allows teams to prioritize alerts and spend time on the more critical problems first.

Each row has a View triage reasoning icon that opens a drawer with the agent's alert description, rationale, and confidence badges.

Alert detail pages now feature a first-class Triage agent section alongside the Troubleshooting agent, embedding the triage experience directly in the alert workflow.

The PR Agent now shows exactly how it calculates risk, broken down by category. This lets you customize how each factor is weighted to match your team's priorities.

The CI Agent now consumes the PR Agent's risk score directly, and introduces full policy management in the Monte Carlo UI: teams can set an installation-wide default gate threshold and add per-repo exception rules, with policies resolving in cascade order from your CI workflow configuration down through per-repo and installation-level UI settings.

To configure: Head to Monte Carlo Settings > GitHub Integration to access the new unified configuration page and adjust thresholds to fit your workflow.

Full documentation: https://docs.getmontecarlo.com/docs/github#risk-factor-weights

You can now control which of your audience recipient channels receive monitor change notifications — opt-in to these notifications via the UI, Notifications-as-Code (using the receive_monitor_changes flag), or the API. (Disabled by default)

When enabled, you'll be notified when a monitor using this audience is disabled, enabled, or deleted. This gives teams full control over which monitor lifecycle activity lands in their channel — so you only see the changes that matter to your workflow.

Learn more here:

• https://docs.getmontecarlo.com/docs/notifications-v2#monitor-change-notifications
  • Notifications-as-code: https://docs.getmontecarlo.com/docs/notifications-as-code-20
  • API: https://apidocs.getmontecarlo.com/#definition-CreateOrUpdateNotificationSetting

You can now subscribe to your Data Operations dashboard and receive a recurring email digest, so your team stays on top of data health without having to log in.

Subscribers get a scheduled summary of dashboard state on a cadence you choose, making it easy to forward to teammates or bring into a recurring meeting.

The new Guides & Resources area behind the question mark icon in the navigation footer has a new layout and expanded content.

From one place your team can access the MC Support agent, enroll in product certification, and keep up with the latest feature announcements. Other resource links that used to live in the Account menu have also moved in here.

Metric monitors now support drift detection and cardinality tracking for ML models. Monitor distribution shifts on the input and output tables of your ML models using PSI, KS Test, or JS Divergence, and surface new or missing distinct values with dedicated cardinality metrics.

Drift metrics compare recent data against a configurable baseline so your team can catch distribution shifts over time. Cardinality metrics flag when new values appear or existing ones go missing in key fields.

These join the existing regression and classification metrics in the metric selector, giving your team a single place to monitor data pipelines, AI, and ML.

Learn more here: https://docs.getmontecarlo.com/docs/available-metrics#ml-metrics

Three new skills in the MC Agent Toolkit help your coding agent tackle warehouse optimization and agent monitoring.

Cost Savings identifies wasteful assets and helps clean your warehouse of stale and unneeded tables. Performance Diagnosis surfaces your worst-performing jobs and provides guidance on improving them. Agent Monitoring explores your agents and sets up monitors across all four supported types.

For best results, connect a database connector alongside the MC Agent Toolkit plugin (v1.4.0 or later).

Three new skills in the MC Agent Toolkit bring the full alert lifecycle into your coding agent.

Automated Triage fetches recent alerts, scores each one by confidence and impact, runs deep troubleshooting on high-signal alerts, and recommends actions. Start from a built-in example workflow or customize to match how your team responds.

Root Cause Analysis systematically investigates data incidents across freshness, volume, schema, field metrics, and ETL failures. It walks the lineage chain upstream, checks ETL jobs across Airflow, dbt, and Databricks, detects query changes, and profiles actual data when a database connector is available.

Remediation picks up where investigation ends. It discovers available tools, proposes a fix with risk assessment and rollback plan, executes with safety rails, and documents everything on the alert. All three skills compose naturally: triage surfaces alerts, root cause analysis investigates them, and remediation fixes them.

The Troubleshooting Agent now investigates job failures as a first-class incident type — in addition to the freshness, volume, and schema incidents it already handles.

What's new:

• Azure Data Factory pipeline failures — when an ADF pipeline failure causes a downstream data incident, the agent now identifies the exact failing pipeline in its root cause analysis, rather than flagging a pipeline-related issue
• Databricks Workflow failures are also supported
• Job runtime context across all incident types — even when an incident isn't a job failure itself, the agent can now correlate it with a long-running or failed job upstream (e.g. a freshness anomaly caused by a slow ADF or Databricks job)