Python Openstackmcp Server

What openstack-kr/python-openstackmcp-server MCP server does

The openstack-kr/python-openstackmcp-server MCP server exposes OpenStack infrastructure operations as MCP tools for AI assistants. It is aimed at environments where an assistant needs to inspect or change cloud resources rather than only answer questions about them.

The available operations include listing, retrieving, creating, updating, and deleting compute servers, depending on the resource and tool. Server tools also support actions, attaching and detaching volumes, and updating a server's name, hostname, or description. Flavor tools list the available server hardware configurations.

Image tools retrieve individual images, list images with optional filters, create images, and delete images. Image creation can use a volume or an OpenStack image import method. Supported import methods include glance-direct, web-download, copy-image, and glance-download; some methods require additional image location or source details.

Identity-related tools cover regions, domains, and projects. For each of these resource types, the material lists retrieval and collection operations plus create, update, and delete operations where applicable. Network support includes listing networks with optional filtering and creating networks.

How it works

An MCP-compatible AI client communicates with the server using the Model Context Protocol. The server uses openstacksdk to communicate with the OpenStack cloud through its REST API. This separates the assistant-facing tool interface from the cloud service calls.

The server reads OpenStack connection details from a clouds.yaml file. The documented configuration includes an authentication URL, username, password, project name, domain settings, region, interface, and Identity API version. The selected cloud configuration is supplied through the OS_CLIENT_CONFIG_FILE environment variable.

Operations that change infrastructure should be treated as administrative actions: creating or deleting servers, images, regions, domains, or projects can alter the cloud state. The available tools make those operations callable by the connected MCP client, subject to the permissions of the configured OpenStack credentials.

Setup and configuration

The documented setup requires Python 3.10 or newer, OpenStack credentials in clouds.yaml, and an installed Claude Desktop client. Configure a cloud entry with the authentication and region values appropriate for the target OpenStack deployment.

Claude Desktop can launch the server with either a Python command or uvx. The documented uvx configuration uses the python-openstackmcp-server package and sets OS_CLIENT_CONFIG_FILE to the path of the clouds.yaml file. The Python example uses the same package as an argument to the configured Python executable.

The Claude Desktop configuration file location depends on the operating system:

• macOS: $HOME/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\\Claude\\Claude_desktop_config.json
• Linux: $HOME/.config/Claude/claude_desktop_config.json

Replace the example paths and credential values with paths and values from the local environment before starting the client.

Limitations and notes

The provided material documents a focused set of OpenStack services rather than every OpenStack API. In particular, the listed tools cover compute, images, identity resources, networks, and volume attachment operations. It does not establish support for other service families or for every operation within those services.

Image filtering is optional: all available images are returned when no filter values are provided, and filters are applied only when the caller supplies them. Image imports also have method-specific requirements. Web downloads need a reachable image URI, while glance-download needs a Glance region and image ID; the import_options field is required for image creation without a volume.

The openstack-kr/python-openstackmcp-server MCP server depends on valid OpenStack credentials and permissions. The README does not describe hosted access, OAuth, a web endpoint, or a separate managed service. It identifies the project as licensed under Apache License 2.0.