wolfdancer/rockfish-mcp

Rockfish MCP Server

A Model Context Protocol (MCP) server that provides access to the Rockfish API, enabling AI assistants to interact with Rockfish's machine learning platform.

Features

This MCP server provides tools for the following Rockfish resources:

• Databases: Create, list, update, and delete databases
• Worker Sets: Manage worker sets for distributed processing
• Workflows: Create and manage ML workflows
• Models: Upload, list, and manage ML models
• Projects: Organize and manage projects
• Datasets: Create and manage datasets

Installation

1. Clone the repository and set up your virtual environment (Python 3.12 or below):

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

2. Install dependencies (choose one method):

Method A: Install with dev tools (recommended for contributors):

pip install -e ".[dev]"

Method B: Install from requirements.txt (exact locked versions):

pip install -r requirements.txt

Method C: Install runtime only (for production):

pip install -e .

3. Set up environment variables:

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

Configuration

Create a .env file with your Rockfish API credentials:

ROCKFISH_API_KEY=your_api_key_here
ROCKFISH_API_URL=https://api.rockfish.ai

If you want to use a specific Rockfish Organization and/or Rockfish Project, add the following to the .env file too:

ROCKFISH_ORGANIZATION_ID=your_organization_id_here
ROCKFISH_PROJECT_ID=your_project_id_here

Usage

Run the MCP server:

python -m rockfish_mcp.server

Or use the console script:

rockfish-mcp

Claude Desktop Setup

To use this MCP server with Claude Desktop:

1. Complete some of the installation steps above (clone, install dependencies). Note that you do not need to start the MCP server manually for using it with Claude Desktop. Claude Desktop will automatically start it for you when you follow the steps below. Also, the .env file doesn't need to be created, we will be adding the environment variables to the Claude setup.

2. Find your Claude Desktop configuration directory:

   • macOS: ~/Library/Application Support/Claude/
   • Windows: %APPDATA%\Claude\
   • Linux: ~/.config/Claude/

3. Create or edit the claude_desktop_config.json file in that directory:

Note that setting ROCKFISH_ORGANIZATION_ID and ROCKFISH_PROJECT_ID is optional. If you don't set these variables, the default organization and/or default project will be used.

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

Note: Environment variables (API keys, URLs, organization/project IDs) are configured in the .env file in the project root, not in claude_desktop_config.json.

4. Update the paths in the configuration:

   • Replace /path/to/your/project/.venv/bin/python with the actual path to your Python executable
   • All API keys and URLs should be configured in the .env file (see step 5)

5. Get the correct Python path by running this command in your project directory:

which python

6. Example configuration (replace with your actual path):

{
  "mcpServers": {
    "rockfish": {
      "command": "/Users/shane/code/rockfish-mcp/.venv/bin/python",
      "args": ["-m", "rockfish_mcp.server"]
    }
  }
}

Configure your API key and URL in the .env file:

ROCKFISH_API_KEY=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9...
ROCKFISH_API_URL=https://api.rockfish.ai

7. Restart Claude Desktop after making these changes

8. Test the connection by asking Claude to list your Rockfish databases or projects

MCP Inspector Setup

The MCP Inspector is a debugging tool that helps you test your MCP server before connecting it to Claude Desktop.

Installation

npx @modelcontextprotocol/inspector

Usage

1. Start the MCP Inspector:

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

2. Or create a test script for easier repeated testing:

#!/bin/bash
# test-mcp.sh
export ROCKFISH_API_KEY="your_api_key_here"
export ROCKFISH_API_URL="https://api.rockfish.ai"
npx @modelcontextprotocol/inspector /Users/shane/code/rockfish-mcp/.venv/bin/python -m rockfish_mcp.server

Make it executable and run:

chmod +x test-mcp.sh
./test-mcp.sh

3. The Inspector will open in your browser and show:

   • Available tools (should show all 32 Rockfish tools)
   • Tool schemas and descriptions
   • Interactive tool testing interface

4. Test your tools by:

   • Selecting a tool from the list (e.g., list_databases)
   • Filling in required parameters
   • Clicking "Call Tool" to test the API call
   • Viewing the response

Useful Tools to Test First

• list_databases - Simple GET request with no parameters
• list_projects - Another simple list operation
• get_database - Test with a database ID from the list
• create_database - Test creating a new resource

Troubleshooting

• MCP server not appearing: Check that the Python path is correct and the virtual environment is activated
• Authentication errors: Verify your ROCKFISH_API_KEY is correct
• Connection issues: Confirm your ROCKFISH_API_URL is accessible
• Path issues on Windows: Use forward slashes or escaped backslashes in JSON paths

Available Tools

Database Tools

• list_databases: List all databases
• create_database: Create a new database
• get_database: Get a specific database by ID
• update_database: Update a database
• delete_database: Delete a database

Worker Set Tools

• list_worker_sets: List all worker sets
• create_worker_set: Create a new worker set
• get_worker_set: Get a specific worker set by ID
• delete_worker_set: Delete a worker set
• get_worker_set_actions: List actions that the specific worker set can run
• list_available_actions: List all actions available to the user (across all worker sets)

Workflow Tools

• list_workflows: List all workflows
• create_workflow: Create and run a new workflow
• get_workflow: Get a specific workflow by ID
• update_workflow: Update a workflow

Model Tools

• list_models: List all models
• upload_model: Upload a new model
• get_model: Get a specific model by ID
• delete_model: Delete a model

Organization Tools

• get_active_organization: Get the currently active organization
• list_projects: List all organizations

Project Tools

• get_active_project: Get the currently active project
• list_projects: List all projects
• create_project: Create a new project
• get_project: Get a specific project by ID
• update_project: Update a project

Dataset Tools

• list_datasets: List all datasets
• create_dataset: Create a new dataset
• get_dataset: Get a specific dataset by ID
• update_dataset: Update a dataset
• delete_dataset: Delete a dataset
• get_dataset_schema: Get dataset metadata present in its schema

SDK Tools (Synthetic Data Generation)

These tools use the Rockfish Python SDK to provide end-to-end synthetic data generation workflows:

Configuration Management:

• obtain_train_config: Generate training configuration for TabGAN model with automatic column type detection
• update_train_config [experimental]: Modify training hyperparameters or field classifications

Workflow Execution:

• start_training_workflow: Start TabGAN training workflow using cached configuration
• get_workflow_logs: Stream workflow logs with configurable level (DEBUG/INFO/WARN/ERROR) and timeout
• get_trained_model_id: Extract trained model ID from completed training workflow

Synthetic Data Generation:

• start_generation_workflow: Start generation workflow from trained model
• obtain_synthetic_dataset_id: Extract generated dataset ID from completed workflow

Quality Assessment:

• plot_distribution: Generate distribution plots comparing real and synthetic data
• get_marginal_distribution_score: Calculate similarity score between real and synthetic data

Development

Setup

To contribute to this project, install with dev dependencies:

pip install -e ".[dev]"

This installs black, isort, and pytest for code formatting and testing.

Code Formatting

This project uses black and isort to maintain consistent code style.

Format your code before committing:

# Format imports
isort src/rockfish_mcp/

# Format code style
black src/rockfish_mcp/

Check formatting without modifying files:

isort --check-only src/rockfish_mcp/
black --check src/rockfish_mcp/

Running Tests

The project includes automated tests for the Manta incident injection tools. To run tests, you'll need to set up the required environment variables:

Required Environment Variables:

# Add these to your .env file or export them
ROCKFISH_API_KEY=your_api_key_here
ROCKFISH_API_URL=https://api.rockfish.ai
MANTA_API_URL=https://manta.rockfish.ai
ROCKFISH_ORGANIZATION_ID=your_organization_id
ROCKFISH_PROJECT_ID=your_project_id
INCIDENT_CREATION_TEST_DATASET=your_test_dataset_id

Running the tests:

# Run all tests
pytest tests/

# Run specific test file
pytest tests/test_create_incident_dataset.py

# Run tests with verbose output
pytest -v tests/

# Run tests and see print output
pytest -s tests/

Test Coverage:

• test_create_incident_dataset.py: Tests the create_incident_dataset tool with all four incident types:

  • instantaneous-spike-data: Sudden spikes in time-series data
  • sustained-magnitude-change-data: Sustained changes over a time period
  • data-outage-data: Simulated data gaps/outages
  • value-ramp-data: Gradual ramping changes

  Test configurations are defined in

Note: The INCIDENT_CREATION_TEST_DATASET should be a valid dataset ID in your Rockfish project with time-series data suitable for incident injection testing.

Contributing

To contribute to this project:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Format your code with isort and black
5. Add tests if applicable
6. Submit a pull request

License

MIT License