Business planning MCP¶

This business application template provides a comprehensive set of MCP (Model Context Protocol) tools and a customizable UI. The template exposes reusable, domain-specific tools through an MCP server, allowing agents to perform complex operations including data access, analysis, visualization, prediction, and scenario planning either through the provided UI or your own consumption layer.

These tools help business planners and analysts streamline their planning processes by automating data preparation, enabling rapid visual analytics, and facilitating predictive modeling and what-if analyses. The solution addresses key challenges in agentic workflows by implementing a pass-by-reference architecture for artifacts, ensuring that large datasets and visualizations are managed efficiently without overwhelming agent context windows. Through its panel-based artifact system, users can view, share, and track the lineage of all generated outputs, from raw data queries to final predictions and visualizations. The template is designed to work seamlessly with popular MCP-compliant clients, making it accessible to users regardless of their preferred AI assistant interface, while remaining fully customizable to adapt to specific organizational needs and business requirements.

Key features¶

• Seamless data access: Connect to and query data from connected datastores (Snowflake, Databricks, and others) through DataRobot data connectors, enabling agents to access and retrieve relevant business data for analysis.

• Intelligent data preparation: Manipulate, analyze, and transform data using Python-based tools, allowing agents to prepare datasets for modeling without manual intervention.

• Interactive visual analytics: Create charts and visualizations from data using flexible Python execution, enabling agents to generate visual insights on demand and share them as shareable artifacts.

• Automated predictions: Make predictions using existing ML deployments in DataRobot, with agents handling the complexity of data formatting and deployment requirements automatically.

• Prediction analysis and comparison: View and analyze the latest predictions from ML deployments, compare different prediction scenarios, and track prediction history to understand model performance over time.

• What-if scenario planning: Perform what-if analyses using ML deployments, allowing planners to explore different business scenarios and understand the impact of variable changes on predictions.

• Artifact management and lineage: View, share, and inspect artifacts (datasets, charts) created by agents, with full provenance tracking to understand how artifacts were generated and their upstream dependencies.

• MCP client compatibility: Access capabilities from MCP-compliant clients such as Claude Desktop, Cursor, and SAP Joule, providing flexibility in how users interact with the business planning agent.

• Customizable toolset: Customize tool implementations and docstrings to adapt the agent's behavior to specific business requirements and use cases.

Architecture¶

This template provides an end-to-end architecture for agentic business planning workflows, from data access through artifact creation and sharing. The solution consists of an MCP server that exposes planning tools, a panel client for managing intermediate artifacts, and a panel viewer for displaying and sharing results. The architecture uses a pass-by-reference approach for artifacts, returning pointers to artifacts rather than embedding large data directly in agent context, ensuring reliable agent performance while maintaining the ability to view and share detailed results through a hosted panel viewer application.