Skip to content

datarobot-mcp

Repository: github.com/datarobot-community/af-component-datarobot-mcp

Adds a FastMCP server to your recipe, deployed as a DataRobot Custom Application. The component provides a ready-to-deploy MCP server with a comprehensive set of DataRobot predictive tools (projects, models, deployments, predictions) and optional integrations for Google Drive, Jira, Confluence, and Microsoft Graph via DataRobot OAuth providers.

Like fastapi-backend, this component is repeatable — apply it multiple times with different mcp_app_name values to run multiple independent MCP backends from a single recipe.

Prerequisites

  • Python 3.11+
  • uv installed
  • dr installed
  • The base component already applied

Apply

dr component add https://github.com/datarobot-community/af-component-datarobot-mcp .

Or with copier directly:

uvx copier copy datarobot-community/af-component-datarobot-mcp .

The wizard asks for an mcp_app_name (e.g., mcp). This scopes all generated files and the answers file, allowing multiple instances in the same project.

Component dependencies

Component Required
base Yes

What it adds

MCP_APP_NAME/
├── MCP_APP_NAME/   # MCP server source (tools, server entrypoint)
└── dev.md            # Local development guide and OAuth setup
infra/infra/MCP_APP_NAME.py   # Pulumi deployment resources
.datarobot/answers/drmcp-MCP_APP_NAME.yml

Available tools

DataRobot predictive tools

Category Tools
Data management Upload datasets to AI Catalog, list catalog items.
Deployment information Get deployment information, generate prediction templates, and validate data.
Deployment management List deployments, get model information, and deploy a model.
Model management Get best model, score datasets, list models.
Predictions Batch predictions (file or AI Catalog), real-time predictions, time series.
Project management List projects, get project datasets.
Training & analysis Analyze datasets, suggest use cases, start Autopilot, ROC curve, feature impact, lift chart.

Integration tools (requires OAuth)

Platform Tools
Google Drive Read files, manage access.
Jira Search, get, create, update, and transition issues.
Confluence Get, create, update pages; add comments; search.
Microsoft Graph Search SharePoint and OneDrive content.

Integration tools require OAuth providers configured in DataRobot. See the generated dev.md for setup instructions.

Local development

Run the MCP server locally:

uv run python -m MCP_APP_NAME

Deploy

dr task deploy

Update

dr component update .datarobot/answers/drmcp-MCP_APP_NAME.yml

Or with copier directly:

uvx copier update -a .datarobot/answers/drmcp-MCP_APP_NAME.yml -A

To update all instances at once:

uvx copier update -a .datarobot/answers/drmcp-*.yml -A

Troubleshooting

dr command not found

Ensure the DataRobot CLI is installed and on your PATH. See the CLI docs for installation instructions.

Copier prompts fail or produce unexpected output

Confirm you are running uv 0.4+ and that copier resolves via uvx. Run uvx copier --version to verify.

Integration tools return auth errors

OAuth providers must be configured in DataRobot before integration tools work. See the generated dev.md inside the template directory.

Multiple instances conflict

Each instance must use a unique mcp_app_name. If two instances share a name, their answers files and generated directories collide. Check .datarobot/answers/ — each instance should have its own drmcp-NAME.yml.