mcp-argo

MCP Argo Server is an MCP-compliant gateway that wraps Argo Workflows with a JSON-RPC interface over STDIN/STDOUT. Built in Go, it leverages Foxy Contexts for RPC handling and interacts with Kubernetes and Argo resources via client-go. The server exposes tools to launch Argo workflows, query their status, and retrieve results, enabling you to programmatically drive workflows through standard MCP tooling and JSON-RPC commands. The included Python test client demonstrates interacting with the server by submitting workflows, polling for status, and collecting results, making it easy to integrate Argo-driven automation into your pipelines.

To use it, start the MCP Argo server using the project’s development workflow (via the dev container) and then connect to it using the provided RPC tools or the Python test client. The README describes how to launch a sample workflow, monitor its progress, and inspect results. You can also inspect the Argo server’s UI at the local URL shown in the README to verify workflows are created and completed as expected.

Prerequisites:

• Docker and/or a development container environment (the project is configured to run inside a dev container).
• Git and Go installed if you intend to run locally outside the dev container.
• kubectl and k3d for cluster provisioning as described in the project.

Install and setup steps:

1. Clone the repository: git clone <repository-url> cd <repository-directory>

2. Open the project in a development container (the README indicates this is the recommended approach). If using VS Code, use Remote - Containers to open the folder so dependencies are pre-installed.

3. If you prefer to run locally (outside the dev container):

   • Ensure Go modules are initialized and dependencies are tidy: go mod tidy
   • You may need to install Go if it isn't already installed on your machine.

4. Build/run the server (as per the project’s Makefile):

   • Start the development cluster and Argo: make cluster
   • Build and run the MCP Argo server: make run

5. Verify the environment:

   • Check the Kubernetes cluster information: kubectl cluster-info
   • Access the Argo UI at the URL shown in the README (https://localhost:2746/workflows/argo/).

Note: The project is designed to work inside a development container for a seamless setup. The steps above align with the references in the README for tidying modules, starting the cluster, and running tests.

Tips and considerations:

• The server exposes a Python test client (python/test_with_autogen.py) that demonstrates workflow submission, status polling, and result retrieval. Run it from the python directory after installing dependencies (make install).
• To run tests, ensure you have Python dependencies installed in the python directory, then execute the test script to observe the MCP Argo server in action.
• The Argo UI is accessible locally; use it to visually verify workflow submissions and their statuses while developing or debugging.
• Since the project relies on a development container for dependencies, make sure to use the same environment when reproducing issues or collaborating with others to avoid environment drift.
• If you need to customize environment variables (e.g., for Kubernetes or Argo endpoints), extend the mcp_config env map with VAR_NAME: "description" placeholders or actual values as needed.