Best Practices
-
Create a report in the Recast UI using the same inputs and cross check values that the MCP reports when using the MCP for high stakes reporting.
-
Make sure the MCP reports all the input parameters it used when running reports and well as which report type it used to answer your question.
MCP vs API
The MCP works by calling the Recast APIs. However there are some cases where the MCP is right tool vs the APIs.
|
When you want to… |
Use |
|---|---|
|
Schedule a script to run every day |
APIs |
|
Ask a one off question about your historical impact |
MCP |
|
Plan future budget |
APIs |
|
Pull values from Recast and apply an additional calculation on top |
APIs |
|
Create a Recast report to share |
MCP |
Common Mistakes
-
Forecasting, Optimizing or Planning via MCP: The MCP currently only has access to the Reporter APIs and the deployments API which means it’s scope is limited to creating and reading reports given a deployment id.
-
Allowing the MCP to select deployments: Make sure to include the KPI you want to get your results for in your request to the MCP
-
Lacking specificity: The more precise you can be with your request, the more accurate the MCP’s response will be. Specify whether you would like the MCP to report direct or total results and mean or median values where relevant.
Escalating MCP questions to Recast Staff
To help you understand your results from the MCP, we need the following details:
-
The the exact prompt you gave your AI agent
-
The reporter inputs used (report type, deployment, dates, channels). You should be able to fetch additional details about the steps your AI agent took when answering your question via MCP by opening up the collapsed ‘tool call details’ in your chat log. Provide as much of this when reporting questions.
-
The report ID that the agent created
-
An export of the chat/ the artifact created by your agent