• TRIAGE
  • ANALYSIS
  • REPAIR
  • VERIFY

How to troubleshoot failed pipeline runs with your AI

Connect Dagster to your agent to pull error logs and run history directly into your development environment.

Ask AI about this page

Short answer

How can I find and fix Dagster errors using my AI?

Your agent uses the Dagster MCP to scan for failed runs and fetch the specific logs where the error occurred. This lets you debug pipeline failures without switching between your IDE and the Dagster UI. You get a direct path from a failed job to the specific traceback.

Where your debugging efforts land.

Debugging outcomes.

The agent organizes the data from your Dagster instance into actionable outcomes.

TRIAGE

Identify failed jobs

The agent scans your recent history to find exactly which runs didn't complete. You get a clear list of what broke.

ANALYSIS

Extract error logs

The agent pulls the specific execution logs for a failed run. You see the traceback and error messages in your chat.

REPAIR

Human-led fixes

The agent identifies the root cause, but you decide how to patch the code or adjust the pipeline. You keep full control over the fix.

VERIFY

Confirm run status

After you apply a fix, the agent checks the status of subsequent runs to ensure the issue is resolved.

The workflow

What your AI does when a run fails.

The agent handles the heavy lifting of navigating the Dagster API so you can focus on the code.

  1. Scan for failures

    The agent looks through your recent execution history to find any runs that ended in an error state.

    list_runs

  2. Isolate the error

    Once a failed run is identified, the agent pulls the detailed execution data for that specific instance.

    get_run

  3. Summarize the traceback

    The agent parses the logs to find the specific exception or error message that caused the crash.

    get_run

  4. Propose a solution

    The agent compares the error log against your current code to suggest what needs to change.

    get_run

Try it

Copy these to start.

Use these prompts to kick off the troubleshooting process.

Starting points

These are opening lines. Swap out the specific run IDs or job names with your actual Dagster data.

Dagster Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_9Cla5bbOzNXXdswxQ5yhHdcbDLhh8oRbE80hiC24/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Dagster capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "dagster-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_9Cla5bbOzNXXdswxQ5yhHdcbDLhh8oRbE80hiC24/mcp"
    }
  }
}
Copy into chat04

  • Show me all the failed Dagster runs from the last hour.

  • Get the logs for run ID 'run-123' and tell me why it failed.

  • Check the status of the most recent 'etl_pipeline' run.

  • Find the error message in the logs for the failed 'data_sync' job.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Dagster once, then ask.

One connection links your Dagster instance to your agent. Your credentials stay encrypted, and you can start debugging immediately.

Connect Dagster to your AI

FAQ

How the Dagster MCP behaves

  • 01

    Can the AI trigger a new run?

    No. The agent can only read run history and fetch logs using the available tools.

  • 02

    Can the AI change my pipeline code?

    No. The agent identifies errors in the logs, but you must manually apply any code changes in your IDE.

  • 03

    Does the agent see all my Dagster data?

    The agent only sees the data it can access through the list and get tools, such as run history and execution logs.

  • 04

    Can the AI delete failed runs?

    No. The agent does not have any tools to delete or modify run history.

  • 05

    How does the agent know which run failed?

    It uses the list_runs tool to check the status of recent executions.

  • More questions about Dagster? The Connector page answers them. See everything the Dagster Connector can do

Connect Dagster to Claude, Cursor, ChatGPT & more