Gridscale MCP
@fabiante
MCP Server to interact with the Gridscale API
Overview
What is Gridscale MCP?
Gridscale MCP is a Model Context Protocol server that lets AI models interact with the Gridscale API, enabling LLM-driven infrastructure provisioning in a Gridscale project. It is intended for use with MCP-compatible clients such as 5ire.
How to use Gridscale MCP?
Build the binary from source with make build, then install it to your $PATH with make install. Run the server with --user-key and --user-token arguments. Configure it as a tool in your MCP client — for example, in 5ire, add a JSON entry with the command gridscale-mcp and the two arguments.
Key features of Gridscale MCP
- create_ip: Provision a new IP address.
- delete_ip: Delete an existing IP address.
- create_storage: Create a storage (partially supported; not all parameters available).
- get_storage_templates: List available storage templates.
- Missing tools (support wanted):
create_serverandcreate_networkare not yet implemented.
Use cases of Gridscale MCP
- Let an LLM allocate and release IP addresses in a Gridscale project.
- Allow an AI assistant to create storage volumes with limited configuration.
- Retrieve storage templates to select appropriate images for provisioning.
- Experiment with automated infrastructure orchestration inside a sandboxed Gridscale project.
FAQ from Gridscale MCP
What does Gridscale MCP do that a direct API client doesn’t?
It bridges the Gridscale API with LLM interfaces via the Model Context Protocol, allowing AI models to manage resources conversationally or programmatically without writing raw API calls.
What are the runtime dependencies for Gridscale MCP?
You need a Go environment to build the binary. Once built, it runs as a standalone executable with no additional runtime dependencies.
Where does my data live when using Gridscale MCP?
All operations are performed against the Gridscale public API. The server itself does not store any data locally; credentials and API interactions are handled through the user-provided key and token.
Are there known limitations?
Yes. Only create_ip, delete_ip, create_storage (partial support), and get_storage_templates are currently functional. create_server and create_network are listed as wanted but not yet implemented.
How does authentication work?
You must pass your Gridscale user key and user token as command-line arguments (--user-key and --user-token). The server uses these credentials to authenticate with the Gridscale API.