Concerns about MCP client/transport coupling – possible refactor?

[faizansf]

I noticed that the MCP section may benefit from some refactoring. Currently, it seems to break some basic SOLID principles—for instance, the MCPClient is instantiated directly within the MCPConnector class. Additionally, when working with the MCPClient class, it’s difficult to create a custom transport (e.g., an HTTP transport), because the StdioTransport is also being instantiated internally, despite the existence of an interface i-e McpTransportInterface.

I'm curious whether these design choices were intentional, or if there's openness to refactoring this for improved flexibility and adherence to design principles?

[ilvalerione]

It was intentional in the sense of keep the McpConnector instance creation focused only on configuration. As you can see here:

https://github.com/inspector-apm/neuron-ai/blob/main/src/MCP/McpClient.php#L16

I imagined the choice of the right transport based on the configuration properties. For HTTP transport it should be "url" instead of command, but this check should be extended.

We are always open for proposals.