You shouldn’t have to go away Cursor to construct, deploy, or monitor a production-grade agent. You’ll be able to wire collectively LangChain, a vector DB, a monitoring instrument, and a deployment pipeline your self, however you’ll spend extra time on that plumbing than on the agent itself. DataRobot is the shortcut. It now lives the place you construct, integrating straight into your IDE throughout the coding agent, instrument layer, and mannequin gateway.
Image what this unlocks. A platform engineer at a fintech firm wires up the World MCP, factors their present LangGraph agent at it, and ships a ruled deployment with monitoring and tracing all earlier than lunch, with out touching their agent code.
DataRobot is the pathway to that workflow. 4 items, one per layer of the stack:
Abilities: drop DataRobot experience into any coding agent
datarobot-agent-skills ships Agent Context Protocol folders for the issues builders ask DataRobot to do: mannequin coaching, predictions, deployment, characteristic engineering, monitoring, explainability, information prep. One set up reaches Claude Code, Cursor, Codex, Gemini CLI, Amp, VS Code Copilot, Goose, Letta, Kilo Code, and OpenCode:
npx ai-agent-skills set up datarobot-oss/datarobot-agent-skillsAfter that, “create a buyer churn mission and begin AutoML” works in your IDE with out memorizing the SDK sample. DataRobot can be within the Cursor market for one-click set up.
Use DataRobot from any MCP shopper
The World MCP is auto-deployed to each DataRobot occasion. 5 strains in .cursor/mcp.json and also you’re accomplished:
{
"mcpServers": {
"datarobot-mcp": {
"url": "https://{DATAROBOT_URL}/api/v2/genai/globalmcp/mcp",
"headers": { "Authorization": "Bearer " }
}
}
}
Want customized instruments or inner companies uncovered over MCP? The af-component-datarobot-mcp template is a FastMCP scaffold with @dr_mcp_tool decorators and Pulumi-managed deployment as a Customized Mannequin App. Native on port 8080, manufacturing on DataRobot serverless.
Brokers cease containing instrument code. They ask the server what’s out there and name it once they want it.
That’s the architectural payoff: add or change instruments with out redeploying the agent. See the LangGraph integration sample for the mcp_tools property that auto-converts MCP instruments into LangChain instruments.
Go from spec to ruled deployment — templates, Agent Help, and the LLM Gateway
datarobot-agent-templates supplies scaffolds for CrewAI, LangGraph, and LlamaIndex. Each ships with Pulumi infrastructure, a dev server, OpenTelemetry tracing, and the required plumbing that turns an area agent right into a ruled DataRobot deployment.
Agent Help (dr help) is the design-before-you-code path. It walks via agent specification, generates agent_spec.md, simulates tool-calling so you’ll be able to validate mannequin and power selection with out burning actual LLM calls, then scaffolds towards the templates.
Beneath all of it: the LLM Gateway, an OpenAI-compatible endpoint at {DATAROBOT_URL}/api/v2/genai/llmgw. Brokers written towards the OpenAI Python SDK work as-is. Switching suppliers is a model-string change. Metering, governance, and credentialing occur on the gateway. All 4 interfaces share auth via a documented credential decision order, with separate Private, Software, and Agent API key sorts whenever you want scoped service-to-service calls.
The way it’s composed
In Cursor: set up Abilities, clone a LangGraph template, level your OpenAI shopper on the LLM Gateway, expose instruments through World MCP, run dr job run deploy. Consequence: a ruled DataRobot deployment with monitoring and tracing.
Each functionality out there within the DataRobot UI can be out there to your IDE and CI pipeline, so you’ll be able to select the floor that matches the duty.
