Best Practices for RudderStack MCP Prompt Templates

Understand the best practices for using RudderStack MCP prompt templates effectively and getting accurate responses.

This guide contains some best practices for using the RudderStack MCP prompt templates for effective and accurate responses.

success
These practices help ensure you get accurate, actionable responses by providing the right context and details in your prompts.

Context enhancement techniques

Use these techniques to make your queries more effective and get better responses.

Details to include

Make sure to include the following details in your prompts, as and when applicable:

  • Descriptive resource names: Use full names of the resources you are referring to, as defined in the RudderStack dashboard. For example, Facebook Conversions API destination is more helpful than just Facebook destination.
  • Time frames: Include specific time frames, especially when trying to investigate a particular issue. For example, past 24 hours, since yesterday, last week, etc.
  • Specific symptoms: Include specific symptoms of the issue you are facing. For example, 1,542 failed events, Volume dropped to 0, I am getting 400 status code error, etc.
  • Exact error messages: Copy-paste the exact error messages from the UI or logs.

Mistakes to avoid

This section covers some patterns to avoid that lead to vague or unhelpful responses.

Patterns to avoid

  • “Something is wrong with my destination”
  • “My data isn’t working”
  • “Can you check my setup?”

Be specific instead

  • “My Facebook Conversions API destination is rejecting events since yesterday with the “missing event_source_url parameter” error”.

Progressive refinement strategies

Use this approach to build on the initial response from the AI when it isn’t helpful or misses important context.

The suggestion about [recommendation from AI] doesn't apply because [your situation].
Can you look specifically at [alternative aspect] given that [additional context]?


Questions? Contact us by email or on Slack