rhoai-mcp: Instructions file for Claude Code

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

RHOAI MCP Server is an MCP (Model Context Protocol) server that enables AI agents to interact with Red Hat OpenShift AI (RHOAI) environments. It provides programmatic access to RHOAI features (projects, workbenches, model serving, pipelines, data connections, storage, training) through domain modules, plus workflow prompts that guide agents through multi-step operations.

Build and Development Commands

# Setup development environment
uv sync                          # Install package
make dev                         # Alias for setup

# Run the server locally
uv run rhoai-mcp                 # Default (stdio transport)
uv run rhoai-mcp --transport streamable-http # HTTP transport

# Testing
make test                        # All tests
make test-unit                   # Unit tests only (tests/training)
make test-integration            # Integration tests (tests/integration)

# Code quality
make lint                        # ruff check
make format                      # ruff format + fix
make typecheck                   # mypy
make check                       # lint + typecheck

# Container operations
make build                       # Build container image
make run-http                    # Run with SSE transport
make run-stdio                   # Run with stdio transport
make run-dev                     # Debug logging + dangerous ops enabled

# MCP Evaluation (requires LLM API key, see .env.eval.example)
make eval                        # Run mock-cluster evals
make eval-live                   # Run all evals (including live cluster)
make eval-scenario SCENARIO=cluster_exploration  # Run a single scenario

Architecture

Project Structure

rhoai-mcp/
├── src/
│   └── rhoai_mcp/               # Main package
│       ├── __init__.py
│       ├── __main__.py          # CLI entry point
│       ├── config.py            # Configuration (pydantic-settings)
│       ├── server.py            # FastMCP server
│       ├── hooks.py             # Pluggy hook specifications
│       ├── plugin.py            # Plugin protocol and base class
│       ├── plugin_manager.py    # Plugin lifecycle management
│       ├── clients/             # K8s client abstractions
│       ├── models/              # Shared Pydantic models
│       ├── utils/               # Helper functions
│       ├── domains/             # Domain modules (pure CRUD operations)
│       │   ├── projects/        # Data Science Project management
│       │   ├── notebooks/       # Kubeflow Notebook/Workbench
│       │   ├── inference/       # KServe InferenceService
│       │   ├── pipelines/       # Data Science Pipelines (DSPA)
│       │   ├── connections/     # S3 data connections
│       │   ├── storage/         # PersistentVolumeClaim
│       │   ├── training/        # Kubeflow Training Operator
│       │   ├── model_registry/   # Model Registry integration
│       │   ├── prompts/         # MCP workflow prompts (18 prompts)
│       │   └── registry.py      # Domain plugin registry (9 plugins)
│       └── composites/          # Cross-cutting composite tools
│           ├── cluster/         # Cluster summaries and exploration
│           ├── training/        # Training workflow orchestration
│           ├── meta/            # Tool discovery and guidance
│           └── registry.py      # Composite plugin registry (3 plugins)
├── tests/                       # Test suite
├── docs/                        # Documentation
├── pyproject.toml               # Project configuration
└── Containerfile                # Container build