dremio-mcp

dremio-mcp

20

The Dremio MCP server facilitates seamless integration between Large Language Models and the Dremio platform, enhancing functionality for data analysis and insights. The server supports multiple configuration modes and encourages open source contributions.

Dremio MCP server

Table of Contents

Introduction

This repo provides an Model Context Protocol (MCP) server for easing LLM integration with Dremio. If you are new to MCP and MCP Servers take our Dremio MCP Server course on Dremio University (DremioU). If you are familiar with these concepts already please proceed below.

%%{init:
{
    "themeVariables": {
        "fontFamily": "Inter"
    }
}
}%%

architecture-beta
    group ws(cloud)[Workstation]

    service cf(database)[Config] in ws
    service mcp(server)[Dremio MCP Server] in ws
    service claude(cloud)[Claude Desktop] in ws

    mcp:B <-- T:cf
    claude:R <--> L:mcp


    group dremio(cloud)[Dremio]
    service de(server)[Dremio Engine] in dremio

    mcp:R <--> L:de

Installation

The MCP server runs locally on the machine that runs the LLM frontend (eg Claude). The installation steps are simple

  1. Clone or download this repository.
  2. Install the uv package manager (note that the MCP server requires python 3.11 or later)
  3. Do a sanity check by running the command and validating the output as shown below.
# cd <toplevel git dir> or add `--directory <toplevel git dir>`
# to the command below

$ uv run dremio-mcp-server --help

 Usage: dremio-mcp-server [OPTIONS] COMMAND [ARGS]...

╭─ Options ────────────────────────────────────────────────────────────────────────╮
│ --install-completion            Install completion for the current shell.        │
│ --show-completion               Show completion for the current shell, to copy   │
│                                 it or customize the installation.                │
│ --help                -h        Show this message and exit.                      │
╰──────────────────────────────────────────────────────────────────────────────────╯
╭─ Commands ───────────────────────────────────────────────────────────────────────╮
│ run      Run the DremioAI MCP server                                             │
│ tools    Support for testing tools directly                                      │
│ config   Configuration management                                                │
╰──────────────────────────────────────────────────────────────────────────────────╯

Initial setup

There are two configurations necessary before the MCP server can be invoked.

  1. The server config file: This will cover the details of connecting and communicating with Dremio
  2. The LLM config file: This covers configuring the LLM desktop app (Claude for now) to make it aware of the MCP server

Quick start

The quickest way to do this setup is -

  1. Create the dremio config file as outlined below and be prepared with these values
$ uv run dremio-mcp-server config create dremioai \
    --uri <dremio uri> \
    # the endpoint portion of the URL for your environment
    --pat <dremio pat> \
    # https://docs.dremio.com/current/security/authentication/personal-access-tokens/#using-a-pat
    # optional: add your project ID if setting up for dremio cloud
    # --project-id <dremio project id>

Note: For security purposes, if you don't want the PAT to leak into your shell history file, create a file with your PAT in it and give it as an argument to the dremio config.

Example:

$ uv run dremio-mcp-server config create dremioai \
    --uri <dremio uri> \
    --pat @/path/to/tokenfile \
  1. Download and install Claude Desktop (Claude)

Note: Claude has system requirements, such as node.js, please validate your system requirements with Claude official documentation.

  1. Create the Claude config file using
$ uv run dremio-mcp-server config create claude
  1. Validate the config files using
$ uv run dremio-mcp-server config list --type claude

Default config file: '/Users/..../Library/Application Support/Claude/claude_desktop_config.json' (exists = True)
{
    'globalShortcut': '',
    'mcpServers': {
        'Dremio': {
            'command': '/opt/homebrew/Cellar/uv/0.6.14/bin/uv',
            'args': [
                'run',
                '--directory',
                '...../dremio-mcp',
                'dremio-mcp-server',
                'run'
            ]
        }
    }
}

$ uv run dremio-mcp-server config list --type claude
Default config file: /Users/..../.config/dremioai/config.yaml (exists = True)
dremio:
  enable_experimental: false
  pat: ....
  uri: ....
tools:
  server_mode: FOR_DATA_PATTERNS

You are done!. You can start Claude and start using the MCP server

Demo

Demo

The rest of the documentation below provides details of the config files

Configuration details

MCP server config file

This file is located by default at $HOME/.config/dremioai/config.yaml but can be overriden using the --config-file option at runtime for dremio-mcp-server

Format
# The dremio section contains 3 main things - the URI to connect, PAT to use
# and optionally the project_id if using with Dremio Cloud
dremio:
    uri: https://.... # the Dremio URI
    pat: "@~/ws/tokens/idl.token" # PAT can be put in a file and used here with @ prefix
    # optional
    # project_id: ....

tools:
    server_mode: FOR_DATA_PATTERNS # the serverm

# Optionally the MCP server can also connect and use a prometheus configuration if it
# has been enabled for your Dremio cluster (typically useful for SW installations)
#prometheus:
#uri: ...
#token: ...
Modes

There are 3 modes

  1. FOR_DATA_PATTERNS - the normal mode where MCP server will allow LLM to look at tables and data to allow pattern discovery and other use cases
  2. FOR_SELF - a mode which allows the MCP server to introspect Dremio system, including workload analysis and so on.
  3. FOR_PROMETHEUS - a mode that allow MCP server to connect to your prometheus setup, if one exists, to enhance insights with Dremio related metrics

Multiple modes can be specified with separated by ,

The LLM (Claude) config file

To setup the Claude config file (refer to this as an example) edit the Claude desktop config file

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
  • Windows: %APPDATA%\Claude\claude_desktop_config.json

And then add this section

{
  "globalShortcut": "",
  "mcpServers": {
    "Dremio": {
      "command": "uv",
      "args": [
        "run",
        "--directory", "<toplevel git directory>"
        "dremio-mcp-server",
        "run"
      ]
    }
  }
}

This will pickup the default location of MCP server config file. It can also be passed in the args section above as "--config-file", "<custom config file>" after run

Further Documentation

  1. : Detailed overview of the Dremio MCP server architecture, including component interactions and data flows.

  2. : Comprehensive guide to available tools, including:

    • Tool categories and types
    • Usage examples
    • Development guidelines
    • Integration support
  3. : Complete configuration reference covering:

    • Dremio connection settings
    • Tool configurations
    • Framework integrations
    • Environment variables
  4. Other frameworks: (Experimental) Integrate the same MCP tools with other agentic frameworks like LangChain or BeeAI without modifying the code base.

Additional Information

This repository is intended to be open source software that encourages contributions of any kind, like adding features, reporting issues and contributing fixes. This is not a part of Dremio product support.

Testing

The project uses pytest for testing. To run the tests:

# Run all tests
$ uv run pytest tests

GitHub Actions automatically runs tests on pull requests and pushes to the main branch.

Contributing

Please see our for details on:

  • Setting up your development environment
  • Making contributions
  • Code style guidelines
  • Documentation requirements
  • Running tests