Dell Enterprise Hub

Documentation

Comprehensive guides and documentation to help you start and utilize our platform.

dell-ai MCP server

The dell-ai MCP (Model Context Protocol) server connects compatible AI assistants, such as Claude Desktop and Cursor, to Dell Enterprise Hub. Once connected, your assistant can find validated models, check hardware compatibility, and prepare deployments from your conversation. MCP support is available starting in dell-ai v1.1.0.

What you can do

  • Find models and applications available on Dell Enterprise Hub.
  • Check which Dell platforms and GPU configurations support a model.
  • Generate Docker, Kubernetes, or Helm deployment configurations.
  • Inspect the local system and existing deployments.
  • Deploy models and applications when you enable deployment tools.

Get started

The setup below runs dell-ai on the same computer as your assistant. Your assistant starts the process for you and uses it to access Dell Enterprise Hub tools.

Install and authenticate

Install the optional MCP dependencies in a Python virtual environment. dell-ai requires Python 3.10 or later:

uv venv .venv
source .venv/bin/activate
uv pip install --upgrade "dell-ai[mcp]>=1.1.0"
dell-ai login
dell-ai whoami

If you already have an environment, activate it and run the install command to upgrade dell-ai with MCP support. You can also use pip install --upgrade "dell-ai[mcp]>=1.1.0" in an activated environment.

Sign in with the Hugging Face account you want your assistant to use. The MCP server reuses this login. Your token must have access to the models you want to use; gated models also require accepting the model author's access conditions on the Hugging Face Hub.

Connect your assistant

For clients that use an mcpServers JSON block, such as Claude Desktop and Cursor, add this entry to the client's MCP configuration:

{
  "mcpServers": {
    "dell-ai": {
      "command": "/absolute/path/to/.venv/bin/dell-ai",
      "args": ["mcp", "start"]
    }
  }
}

Replace /absolute/path/to/.venv/bin/dell-ai with the path to your installed executable. On macOS or Linux, run command -v dell-ai in the activated environment to find it. This lets your assistant find the installation even when it starts outside your terminal.

Other clients may use a different configuration format; enter the same executable path and the arguments mcp start in their MCP settings.

Save the configuration and restart or reconnect your assistant. You should see dell-ai listed as a connected MCP server with its tools available. This setup does not require a separate server configuration file or a command left running in your terminal.

Using the server

Ask your assistant to use the Dell Enterprise Hub tools. For example:

  • “Find the models available for xe9680-nvidia-h200 and show their supported GPU configurations.”
  • “Check whether my Hugging Face account can access google/gemma-3-27b-it.”
  • “Generate a Docker snippet for google/gemma-3-27b-it on xe9680-nvidia-h200 with 8 GPUs. Do not deploy it.”
  • “Show me the available applications and the configuration needed to install Open WebUI.”
  • “List my Dell AI deployments.”

Your assistant calls the appropriate tools and brings the results back into the conversation. You can then refine the model choice, review a configuration, or continue with a deployment. Model availability and compatible configurations come from the current Dell Enterprise Hub catalog.

Built-in tools

These tools are available with the default configuration:

Task Tools
Find and inspect models list_models, search_models, get_model, check_model_access
Explore Dell platforms and compatibility list_platforms, get_platform, get_compatible_platforms, get_platform_system_info
Prepare model deployments get_deployment_snippet, get_container_tags, get_goodput_scenarios
Explore applications and generate Helm snippets list_apps, get_app, get_app_snippet
Inspect the server's local system and deployments describe_system, list_deployments
Read available agent skills list_skills, get_skill

Generating a snippet does not execute it. Model snippets can contain the $$_TOKEN_$$ placeholder; replace it with your Hugging Face token before running the snippet manually.

Enable deployments

Deployment tools are disabled by default. To let your assistant deploy and remove workloads, add --allow-destructive to the args array in your client's configuration:

["mcp", "start", "--allow-destructive"]

Reconnect your assistant to make deploy_model, deploy_app, and undeploy available. This option also enables install_skill, which installs agent skills into assistant directories.

Deployments use the Docker, Kubernetes, or Helm tools configured on the computer running dell-ai. For local Docker deployments, this should be the Dell system where you want the model to run. Review the proposed configuration before asking your assistant to deploy it.

For example, after reviewing a generated configuration:

Deploy the model with the configuration we just reviewed, then show the deployment ID and endpoint.

Use list_deployments to identify a deployment before removing it. undeploy handles Docker containers and Kubernetes deployments; Helm releases require the manual helm uninstall command reported by the tool.

Troubleshooting

  • The MCP extra is missing: install or upgrade "dell-ai[mcp]>=1.1.0" in the environment used by the client's executable.
  • The client cannot start the server: check the executable's absolute path in your client's configuration, then reconnect.
  • Deployment tools are missing: add --allow-destructive to the server arguments, then reconnect.
  • Hugging Face access fails: run dell-ai whoami under the same user account as your assistant, and check any gated-model access requirements.

Learn more