Rockfish

You run a Rockfish MCP Server that provides programmatic access to Rockfish’s machine learning platform. It lets you manage databases, worker sets, workflows, models, projects, and datasets from AI assistants or tooling clients, enabling automated ML workflows and synthetic data generation with secure API access.

How to use

Connect your MCP client to the Rockfish MCP Server using the local runtime configuration. You can perform common actions such as listing databases, creating new resources, updating existing ones, and running workflows or model operations. When you start the server, your client can call endpoints to manage Rockfish resources and to trigger synthetic data generation or quality assessments as part of end-to-end ML pipelines.

How to install

Prerequisites: you need Python 3.12 or lower installed on your system.

git clone https://github.com/yourusername/rockfish-mcp.git
cd rockfish-mcp
python3.11 -m venv .venv
source .venv/bin/activate

Install dependencies using one of the following methods.

# Method A: Development installation (recommended for contributors)
pip install -e ".[dev]"

# Method B: Exact locked dependencies
pip install -r requirements.txt

# Method C: Runtime/production installation
pip install -e .

Configure environment variables for Rockfish access. Copy the example and edit with your credentials.

cp .env.example .env
# Edit .env and add your Rockfish API key and URL

Additional setup and usage notes

Prepare your environment file with the credentials and optional organization/project IDs to scope your Rockfish workspace.

ROCKFISH_API_KEY=your_api_key_here
ROCKFISH_API_URL=https://api.rockfish.ai
ROCKFISH_ORGANIZATION_ID=your_organization_id_here
ROCKFISH_PROJECT_ID=your_project_id_here

Starting the MCP server

Run the server from your virtual environment so your MCP client can connect.

python -m rockfish_mcp.server

Using with Claude Desktop

If you plan to use Claude Desktop, you can configure a local MCP server entry. Set the run command to the Python interpreter in your environment and specify the module to launch.

{
  "mcpServers": {
    "rockfish": {
      "command": "/path/to/your/project/.venv/bin/python",
      "args": ["-m", "rockfish_mcp.server"]
    }
  }
}

MCP Inspector setup and testing

Use the MCP Inspector to test your server locally before connecting Claude Desktop. It lets you run tool calls and view responses in a browser-based interface.

npx @modelcontextprotocol/inspector

Start the inspector with the Python runtime and module path to your server.

npx @modelcontextprotocol/inspector /Users/you/code/rockfish-mcp/.venv/bin/python -m rockfish_mcp.server

Useful testing tips with the Inspector

Open the inspector in your browser to see available tools, test parameter inputs, and review responses. Begin by selecting a simple tool like listing databases or projects, provide required inputs, and call the tool to validate your setup.

Available tools

list_databases

Retrieve all databases in the Rockfish workspace

create_database

Create a new database with specified properties

get_database

Fetch a database by its ID

update_database

Modify properties of an existing database

delete_database

Remove a database by its ID

list_worker_sets

List all worker sets available for distributed tasks

create_worker_set

Create a new worker set for parallel processing

get_worker_set

Get details for a specific worker set by ID

delete_worker_set

Delete a worker set by ID

get_worker_set_actions

List actions that a worker set can perform

list_available_actions

List all user-available actions across worker sets

list_workflows

List all workflows in the project

create_workflow

Create and execute a new ML workflow

get_workflow

Get details for a specific workflow by ID

update_workflow

Update an existing workflow's configuration

list_models

List all uploaded ML models

upload_model

Upload a new ML model to the workspace

get_model

Retrieve a model by ID

delete_model

Delete a model by ID

get_active_organization

Get the currently active organization

list_projects

List all projects in the organization

get_active_project

Get the currently active project

create_project

Create a new project

get_project

Get a project by ID

update_project

Update a project's details

list_datasets

List all datasets in the workspace

create_dataset

Create a new dataset

get_dataset

Get a dataset by ID

update_dataset

Update a dataset

delete_dataset

Delete a dataset

get_dataset_schema

Get metadata schema for a dataset

obtain_train_config

Generate training configuration for TabGAN with automatic column type detection

update_train_config

Modify training hyperparameters or field classifications (experimental)

start_training_workflow

Start a training workflow using cached configuration

get_workflow_logs

Stream workflow logs with adjustable level and timeout

get_trained_model_id

Extract trained model ID from a finished training workflow

start_generation_workflow

Start synthetic data generation workflow from a trained model

obtain_synthetic_dataset_id

Extract generated dataset ID from a completed workflow

plot_distribution

Generate distribution plots comparing real and synthetic data

get_marginal_distribution_score

Calculate similarity score between real and synthetic data