Monte Carlo Agent Toolkit

Cortex Code (CoCo) plugin that brings Monte Carlo data observability into Snowflake's CLI coding agent. Bundles skills, hooks, commands, and the Monte Carlo MCP server in a single plugin.

Requires Python 3.10+ and the Cortex Code CLI.

Installation

Cortex Code's cortex plugin install can't consume this plugin directly from the repo, because the skill entries are symlinks shared across all editor plugins (they point outside the plugin directory, which the installer rejects). The install script materializes a symlink-resolved copy and installs that for you:

git clone https://github.com/monte-carlo-data/mc-agent-toolkit.git
cd mc-agent-toolkit
bash plugins/cortex-code/scripts/install.sh

The script clones (or uses your local checkout), resolves the symlinks into a staging copy, and runs cortex plugin install. Verify with:

cortex plugin list

You should see mc-agent-toolkit listed as [enabled, managed] with command, skill, hook, and MCP-server components.

Setup

1. Authenticate with Monte Carlo — run /mcp inside Cortex Code (or cortex mcp from the CLI), select the Monte Carlo server, and complete the OAuth flow.
2. Verify — ask Cortex: "Test my Monte Carlo connection."

Usage

• Invoke a skill explicitly with $<skill-name>, e.g. $monte-carlo-prevent or $monte-carlo-asset-health. Run /skill list to see what's available.
• Auto-activation — Cortex selects the right skill from your request (e.g. editing a dbt model activates Prevent; "how healthy is this table?" activates Asset Health).
• Validation — run /mc-validate (or /mc-agent-toolkit:mc-validate) to generate or run validation queries for the dbt models you changed this session.

Telemetry

The toolkit sends anonymous skill-usage telemetry by default — which skills are invoked, how often. Each event includes an opaque per-install UUID, a per-session UUID, the skill name, the toolkit version, and the editor it runs in (cortex-code). No prompts, arguments, or code are ever sent.

It also sends a Toolkit Installed beacon once per toolkit version — the first time you start Cortex Code after installing, and again after each version change — deduped by a local marker, independent of whether you ever run a skill. It carries the install/session UUIDs, toolkit version, and editor (cortex-code) — no skill field. Like the skill beacon, it is fail-open and non-blocking.

Authenticated MCP traffic (v1.13.3+). The same anonymous install_id, a per-session id, and the toolkit version also ride as HTTP headers (x-mcd-toolkit-install-id, x-mcd-toolkit-session-id, x-mcd-toolkit-version) on authenticated requests to the Monte Carlo MCP server. This lets the otherwise-anonymous install record be correlated with your account's MCP tool usage server-side — still no prompts, arguments, or code, only the opaque ids and version. The opt-out below disables these headers too.

To opt out, set MC_AGENT_TOOLKIT_TELEMETRY_DISABLED=1 in your shell environment before starting Cortex Code. The toolkit will not phone home.

The data is stored in Mixpanel and Datadog and is used only for product-development decisions about which skills to invest in. The UUIDs are generated locally on first session and stored under ~/.snowflake/cortex/mc-agent-toolkit/ (Cortex Code's own config home) — separate from any Claude Code toolkit install, so each editor keeps its own anonymous identity. Deleting that directory resets the Cortex Code install identity to a fresh one.

Available Features

Feature	Description	Details
Monitoring Advisor	Analyzes data coverage, creates monitors for warehouse tables and AI agents — covers coverage gaps, use-case analysis, data monitor creation, and agent observability.	Skill README
Prevent	Gates dbt model edits with impact assessments, generates monitors-as-code, and produces targeted validation queries. Full hook enforcement.	Skill README
Generate Validation Notebook	Generates SQL validation notebooks for dbt model changes from a PR or local repo.	Skill README
Push Ingestion	Generates warehouse-specific collection scripts for pushing metadata, lineage, and query logs to Monte Carlo. Includes 10 /mc-* slash commands.	Skill README
Automated Triage	Guides you through automated alert triage — scoring, deep troubleshooting, classification, and actions. Requires extended MCP toolset.	SKILL

All 17 skills are bundled — see the skills directory for the full list.

The Prevent feature includes PreToolUse hooks that block edits to dbt SQL files until an impact assessment runs. The hooks only fire on .sql files inside dbt model, macro, or snapshot directories — they do not affect non-dbt files.

To disable the Prevent hooks, set MC_PREVENT_HOOKS_DISABLED=1 in your environment. The hooks will exit immediately without blocking any edits.

See the Prevent Hook Behavior section in the plugins README for full details on scope and configuration.

Updating

Re-run the installer from your repo checkout (the clone from the Installation step) to pull the latest skills and hooks:

cd mc-agent-toolkit
git pull
bash plugins/cortex-code/scripts/install.sh

Uninstalling

cortex plugin uninstall mc-agent-toolkit

Architecture

Cortex Code wraps Claude Code, so this plugin reuses the same SKILL.md skills, the same prevent hook lifecycle (PreToolUse/PostToolUse/Stop), and the same MCP configuration. The one Cortex-specific adaptation is the prevent transcript reader: Cortex stores session messages in a sibling <id>.history.jsonl and persists hook output back into the transcript, so the gate scans only assistant-authored text to find its completion marker. See the plugins README for the editor-support comparison and hook-format details, and the Plugin Architecture Guide for the shared-core/adapter design.

Changelog

See CHANGELOG.md.