Deephaven MCP

Deephaven MCP connects Deephaven to your AI development workflow. It uses the Model Context Protocol (MCP), an open-source standard for connecting AI applications to external systems such as databases and filesystems. MCP connects AI applications to services the way USB connects external devices to computers.

With Deephaven MCP, AI models can query tables and run scripts directly in your Deephaven sessions. They can also write Deephaven code for you. The tables the AI builds in your session are ordinary Deephaven tables. When they're built on live data, they keep updating as new data arrives. While AI models do the busy work, you get to focus on turning that work into actionable insights.

Deephaven MCP has three components:

  • Systems server — a server that manages and connects to your Deephaven Community Core and Deephaven Enterprise systems, exposing sessions and tables to your AI tools.
  • Docs server — a public server hosted by Deephaven that answers natural-language questions about Deephaven, using its documentation to return code examples and explanations.
  • dhcli — a command-line tool that configures and operates your Deephaven systems from the shell.

Get started

The Deephaven MCP README on GitHub covers installation, configuration, and usage, including the dhcli command-line tool and setup for AI tools such as Claude Desktop, Cursor, VS Code, and Windsurf.

You don't need a running Deephaven server to get started, because dhcli can start Community Core sessions for you. These sessions use Python by default. To start a Groovy session, run dhcli session create dev --language Groovy --launch-method docker, where dev is the name of the new session. A Community Core session needs a name. Groovy sessions run in Docker, so you need Docker installed.

To use a Deephaven server you run yourself, start one with the Quickstart. Then add it as a session, following the Community Core Quick Start in the README. Pass --language Groovy when you add it.

To get Groovy code in answers from the documentation assistant, pass --language groovy to dhcli docs ask.