The Model Context Protocol (MCP) defines a discovery-based interface that can be accessed over JSON-RPC. It is designed primarily for dynamic, agentic clients capable of real-time tool discovery and adaptation.
Non-agentic clients
While you could write a client that talks to CloudBees Unify using the CloudBees MCP server, CloudBees strongly recommends against doing so.
The CloudBees MCP protocol offers runtime discovery that AI agents use to discover tools, resources, and prompts from the server. This discovery and the use of a large-language-model (tool calling) means that the AI agent can immediately adapt to changes in the names, parameters, and descriptions for tools that can occur between releases.
While this might feel uncomfortable in a classical development world of versioned interfaces and contracts, in the world of tools, it’s normal and expected.
The challenge comes when you try to connect a statically developed client to an MCP tool — the interface can, and likely will, change between releases.