Dagster Expert

Core Dagster Concepts

Brief definitions only (see reference files for detailed examples):

• Asset: Persistent object (table, file, model) produced by your pipeline
• Component: Reusable building block that generates definitions (assets, schedules, sensors, jobs, etc.) relevant to a particular domain.

Integration Workflow

When integrating with ANY external tool or service, read the Integration libraries index. This contains information about which integration libraries exist, and references on how to create new custom integrations for tools that do not have a published library.

Programmatic Access: dg CLI and the Dagster Plus MCP server

There are two ways to interact with Dagster programmatically. Pick by where the work happens:

• The dg CLI — everything in the local project: adding definitions, scaffolding, validating, exploring project structure, and launching runs locally. Installed as part of the dagster-dg-cli package. If a relevant CLI command for a local task exists, always attempt to use it.
• The Dagster Plus MCP server — querying and managing a deployed Dagster Plus organization: runs, assets, deployments, code locations, alert policies, Issues, and insights metrics. When it is connected, prefer its tools over the equivalent dg api commands.

dg api covers the same deployed resources as the MCP server from the command line, and remains the way to reach the parts the server does not expose. Before doing anything against a deployed Dagster Plus environment, read Dagster Plus API: General — it covers how to choose between the two and how to fall back safely.

ONLY explore the existing project structure if it is strictly necessary to accomplish the user's goal. In many cases, existing CLI tools will have sufficient understanding of the project structure, meaning listing and reading existing files is wasteful and unnecessary.

Almost all dg commands that return information have a --json flag that can be used to get the information in a machine-readable format. This should be preferred over the default table output unless you are directly showing the information to the user.

UV Compatibility

Projects typically use uv for dependency management, and it is recommended to use it for dg commands if possible:

uv run dg list defs
uv run dg launch --assets my_asset

CRITICAL: Always Read Reference Files Before Answering

NEVER answer from memory or guess at CLI commands, APIs, or syntax. ALWAYS read the relevant reference file(s) from the Reference Index below before responding.

For every question, identify which reference file(s) are relevant using the index descriptions, read them, then answer based on what you read.

Reference Index

• Asset Selection Syntax — filtering assets by tag, group, kind, upstream, or downstream; AssetSelection in Python, UI search bar, or CLI
• Environment Variables — configuring environment variables across different environments
• Asset Patterns — defining assets, dependencies, metadata, partitions, or multi-asset definitions
• Choosing an Automation Approach — deciding between schedules, sensors, and declarative automation
• Schedules — time-based automation with cron expressions
• Declarative Automation — asset-centric condition-based automation using AutomationCondition
• Asset Sensors — triggering on asset materialization events
• Basic Sensors — event-driven automation with file watching or custom polling
• Run Status Sensors — reacting to run success, failure, or other status changes
• dg check — validating project configuration or definitions
• create-dagster — creating a new Dagster project from scratch
• dg dev — starting a local Dagster development instance
• dg launch — materializing assets or executing jobs locally
• dg list components — seeing available component types for scaffolding
• dg list defs — listing or filtering registered definitions
• Dagster Plus API — dg api or the Dagster Plus MCP server, programmatically querying or managing Dagster Plus resources (assets, runs, deployments, code locations, schedules, sensors, secrets, issues, alert policies, etc.); Dagster credits, compute or warehouse cost, usage, and other insights metrics for an asset, job, or deployment; retrying or re-executing a failed run or backfill; terminating a run
• dg list — exploring project structure (component tree, environment variables, workspace projects)
• Dagster Plus CLI — dg plus, Dagster Plus authentication, configuration, and deployment; logging in, setting config, creating API tokens, deploying code, pulling env vars, managing dbt manifests
• dg scaffold component — creating a custom reusable component type
• dg scaffold defs — adding new definitions (assets, schedules, sensors, components) to a project
• dg utilities — dg utils, inspecting component types, refreshing state-backed component cache
• Creating Components — building a new custom component from scratch
• Designing Component Integrations — designing a component that wraps an external service or tool; custom integrations
• Resolved Framework — defining custom YAML schema types using Resolver, Model, or Resolvable
• Subclassing Components — extending an existing component via subclassing; customize dagster integration component
• Template Variables — using Jinja2 template variables in component YAML (env, dg, context, or custom scopes)
• Creating State-Backed Components — building a component that fetches and caches external state
• Using State-Backed Components — managing state-backed components in production, CI/CD, or refreshing state
• Deployment Configuration Files — build.yaml, container_context.yaml, dagster_cloud.yaml; Dagster Plus deployment configuration; configuring Docker registry, container context, agent queue; Hybrid deployment files
• Integration libraries index for 40+ tools and technologies (dbt, Fivetran, Snowflake, AWS, etc.). — integration, external tool, dagster-*; dbt, fivetran, airbyte, snowflake, bigquery, sling, aws, gcp
• Migration Guides — sensor migration to declarative automation, sensor migration to automation condition